0% found this document useful (0 votes)
31 views78 pages

Chapter4 5

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
31 views78 pages

Chapter4 5

Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 78

Chapter 4

Vectors and Tensors

In this chapter, we will switch our definitions of vectors to be objects that have a length
and direction. Later, we will show that these vectors can be represented as column vectors
in the matrix algebra domain and that the vector operations to be described in this chapter
do have strong correspondence with matrix operations.

We will restrict our discussion to real vectors existing in the three dimensional space,
and we will represent the vectors by an arrow in which the length of the arrow describes its
magnitude and the arrowhead describes its direction. Symbolically, we include an underline
to distinguish the vectors treated in this chapter to those treated in the previous chapter,
e.g. instead of v, we will denote the vector by v.

4.1 Free Vectors and Bound Vectors

There are two types of vectors. The first type is called the bound vector in which the vector
is strongly fixed to a specified point. The other type is called the free vector in which the
vector could be moved in parallel during operations such as addition, dot products or cross
products (see Figure 4.1). Ironically, free vectors are used more prevalently on rigid domains,
while bound vectors are used more predominantly on flowing or flexible domains. Considering
that we will be developing operations on flow fields, we need to insist on using bound vectors
as much as possible. Thus, we base our operations on bound vectors as a default. However,
there will be occasions in which we might change the character of the vector to be free,
specially when using free vectors significantly improves computational efficiency. (When
such change occurs, we will include a special mention.)

109
110 c 2006 Tomas Co, all rights reserved
°

Figure 4.1: Two types of vectors: bound and free.

Figure 4.2: Addition of bound vectors.

4.2 Vector Operations


There are several operations involving vectors. We will begin with the general procedures
and then follow it up with alternative procedures based on unit vectors in the next section.

4.2.1 General Operations


1. Vector Addition.
• Notation: a+b=c
• Procedure: Assuming a and b are bound to the same point, i.e. their tails
coincide at point O, construct a parallelogram with a and b as two adjacent sides.
Then c = a + b is the vector whose tail is at point O and its arrowhead is at the
point directly opposite point O in the parallelogram. (see Figure 4.2).
In case a and b are free vectors, an alternative procedure can be used to obtain the
sum. First, attach the tail of b to the arrowhead of a. Then c is the vector whose
tail coincides with the tail of a and whose arrowhead coincides with arrowhead of
b (see Figure 4.3).
• Property: Vector addition is commutative.
2. Magnitude.
c 2006 Tomas Co, all rights reserved
° 111

Figure 4.3: Addition of free vectors.

Figure 4.4: The norm of a vector.

• Notation: kvk
• Procedure: Measure the length of the arrow (see Figure 4.4).

3. Scalar Product.

• Notation: αv=w
• Procedure: Obtain the magnitude of v and multiply with the scalar factor
α, then w becomes a vector whose length equal αkvk and points in the direction
of v (see Figure 4.5).

4. Dot Product. (also known as Inner Product)

• Notation: a·b=α
• Procedure: Obtain the length of projection of a on the direction of b, i.e.
kak cos(θ), where θ is the included angle between a and b. Then multiply this
value with the length of b. Then α = kakkbk cos(θ) . (see Figure 4.6).
• Property: Inner Products are commutative.

5. Cross Product.

Figure 4.5: Scalar product of a vector.


112 c 2006 Tomas Co, all rights reserved
°

Figure 4.6: Dot product of vector a and b.

Figure 4.7: Cross product of vector a and b.

• Notation: a×b=c
• Procedure: Take the area of a parallelogram formed by a and b joined at the
tails, i.e. area=kakkbk sin(θ), where θ is the angle from a to b in the counterclock-
wise manner. Then c is simply a vector whose magnitude is equal to this area
and points in the direction perpendicular to the plane containing a and b using a
right hand screw system (see Figure 4.7).
• Property: Cross products are anti-commutative, i.e. a × b = −b × a.

4.2.2 Vector Operations based on Unit Vectors

Consider a coordinate system having a basis of three mutually perpendicular axes. Then
unit vectors are simply the vectors of unit length each pointing in the direction of positive
part of the axes. For the Cartesian coordinate system, we will follow the convention shown
in Figure 4.8. Based on the cartesian coordinate system, the unit vectors are then denoted
by δ x , δ y , and δ z .
Based on the general operations described in the previous section, the dot product and
cross products among the unit vectors are given by:
c 2006 Tomas Co, all rights reserved
° 113

Figure 4.8: The Cartesian coordinate system with three axis perpendicular to each other.

Figure 4.9: Vector V as a linear combination of unit vectors.

δx · δx = δy · δy = δz · δz = 1 (4.1)
δx · δy = δx · δz = δy δz = δy · δx = δz · δx = δz δy = 0 (4.2)
δx × δx = δy × δy = δz × δz = 0 (4.3)
δx × δy = δz ; δy × δz = δx ; δz × δx = δy
δy × δx = − δz ; δz × δy = − δx ; δx × δz = − δy (4.4)

Having identified the unit vectors, any vector can now be represented as a linear
combination of the unit vectors as shown in Figure 4.9, i.e.
V = vx δ x + vy δ y + vz δ z (4.5)

The coefficient vx is the magnitude of the projection of V on δ x , i.e. vx = V · δ x .


Likewise, vy = V · δ y and vz = V · δ z .
Based on (4.5) and the identities given in (4.1)-(4.4), the vector operations can be
performed in the following alternative manner:

1. Vector Addition.
v + w = (vx δ x + vy δ y + vz δ z ) + (wx δ x + wy δ y + wz δ z )
= (vx + wx )δ x + (vy + wy )δ y + (vz + wz )δ z
114 c 2006 Tomas Co, all rights reserved
°

2. Norm.

kvk = kvx δ x + vy δ y + vz δ z k
q
= vx2 + vy2 + vz2

3. Scalar Product

α v = α(vx δ x + vy δ y + vz δ z )
= αvx δ x + αvy δ y + αvz δ z

4. Dot Product

v · w = (vx δ x + vy δ y + vz δ z ) · (wx δ x + wy δ y + wz δ z )
= (vx δ x ) · (wx δ x ) + (vx δ x ) · (wy δ y ) + (vx δ x ) · (wz δ z ) +
(vy δ y ) · (wx δ x ) + (vy δ y ) · (wy δ y ) + (vy δ y ) · (wz δ z ) +
(vz δ z ) · (wx δ x ) + (vz δ z ) · (wy δ y ) + (vz δ z ) · (wz δ z )
= (vx wx )(δ x · δ x ) + (vx wy )(δ x · δ y ) + (vx wz )(δ x · δ z ) +
(vy wx )(δ y · δ x ) + (vy wy )(δ y · δ y ) + (vy wz )(δ y · δ z ) +
(vz wx )(δ z · δ x ) + (vz wy )(δ z · δ y ) + (vz wz )(δ z · δ z )
= vx wx + vy wy + vz wz

5. Cross Product

v × w = (vx δ x + vy δ y + vz δ z ) × (wx δ x + wy δ y + wz δ z )
= (vx δ x ) × (wx δ x ) + (vx δ x ) × (wy δ y ) + (vx δ x ) × (wz δ z ) +
(vy δ y ) × (wx δ x ) + (vy δ y ) × (wy δ y ) + (vy δ y ) × (wz δ z ) +
(vz δ z ) × (wx δ x ) + (vz δ z ) × (wy δ y ) + (vz δ z ) × (wz δ z )
= (vx wx )(δ x × δ x ) + (vx wy )(δ x × δ y ) + (vx wz )(δ x × δ z ) +
(vy wx )(δ y × δ x ) + (vy wy )(δ y × δ y ) + (vy wz )(δ y × δ z ) +
(vz wx )(δ z × δ x ) + (vz wy )(δ z × δ y ) + (vz wz )(δ z × δ z )
= (vx wy )δ z − (vx wz )δ y − (vy wx )δ z + (vy wz )δ x
+ (vz wx )δ y − (vz wy )δ x
= (vy wz − vz wy )δ x + (vz wx − vx wz )δ y + (vx wy − vy wx )δ z

As a mnemonic, others have used the determinant formula for the cross product,
 
δx δy δz
v × w = det  vx vy vz  (4.6)
wx wy wz
c 2006 Tomas Co, all rights reserved
° 115

For most people, this formula is easier to remember. Nonetheless, it is important to


remember that (4.6) is just a memory tool and not the definition. (The distinction
between a memory tool and definition is most crucial later when we perform operations
such as “curl” of vector fields.)

4.3 Dyads
Dyads are one type of operators on vectors. The dyad requires two defining vectors as
its arguments, say d1 and d2 , and it is often denoted by writing both defining vectors side
by side, e.g. d1 d2 . The dyad operation consists of a sequence of procedures and is strictly
dependent on the order in which the dyad is written. For instance, d1 d2 is not the same as
d2 d1 . The first vector in the dyad is called the antecedent and the second vector in the
dyad is called the consequent.
Consider the dyad, d1 d2 that is operating on vector v to yield another vector w.
Symbolically, we have (d1 d2 )(v) = w. The procedure of the dyad operators consists of the
following steps: first, one takes the dot product between d2 and v to yield a scalar value
α = d2 · v. Then this value is used for the scalar product (α d1 ) to produce a vector w.
Thus w is a vector whose length is (d2 · v)kd1 k and is pointing in the same direction as d1 .
( Based on this procedure, other use the dot product notation to remind them of the dyad
operation, i.e. (d1 d2 )(V ) = d1 d2 · V . )
Among the different possible dyads, an important set of nine dyads are the unit dyads
formed by the unit vectors:

δxδx, δxδy , δxδz


δy δx, δy δy , δy δz
δz δx δz δy , δz δz

4.4 Tensors
We will define tensors as simply an operator which is a linear combination of all nine unit
dyads.1
τ = τxx δ x δ x + τxy δ x δ y + τxz δ x δ z + . . . + τzz δ z δ z (4.7)

We denote tensors by including double underlines to distinguish it from vectors.

Example 4.1 Stress Tensors


1
More accurately, the tensors defined above is called the second order tensor and vectors are also called
first order tensors. This suggests that there are higher order tensors. However, we will not delve into tensors
that are higher than second order.
116 c 2006 Tomas Co, all rights reserved
°

Figure 4.10: The stress vector with respect to δ z as a sum of the normal stress and shear
stress.

The stress tensor is actually an indicator for the state of stress at a point. Since
stress is defined as force per unit area, the stress at a particular point has meaning
only with respect to a plane. In three dimensional space, the plane S attached to
a point p is automatically defined by a unit normal vector, say n, that is bound
to point p. With respect to plane S, we can determine the amount of stress
along the direction n which is normal to the plane, called the normal stress.
The other stress is directed parallel to the plane, and is called the shear stress.
Taking the sum of the normal stress and the shear stress will simply give the
stress vector at that point, which we can denote as τ (p,n) .
Let T be the stress tensor given by

T = Txx δ x δ x + Txy δ x δ y + Txz δ x δ z


+Tyx δ y δ x + Tyy δ y δ y + Tyz δ y δ z
+Tzx δ z δ x + Tzy δ z δ y + Tzz δ z δ z

To understand the coefficients of T , consider the x–y plane at say point p as


shown in Figure 4.10. The unit normal vector to this plane is δ z . When the
tensor T operates on δ z (based on dyad operations discussed in the previous
section),
τ (p,δz ) = T δ z = Txz δ x + Tyz δ y + Tzz δ z

Since Tzz is the term along δ z , it gives the value of the normal stress to the x–y
plane. On the other hand, the other two terms, Txz and Tyz are simply the x and
y components of the shear stress, acting across the x–y plane.
For the general case,
n = nx δ x + ny δ y + nz δ z
c 2006 Tomas Co, all rights reserved
° 117

Then the stress vector at a point p with respect to the plane specified by a normal
vector n, is obtained by having T operate on n,

τ (p,n) = T n (4.8)
= (Txx δ x δ x + Txy δ x δ y + . . . + Tyz δ y δ z + Tzz δ z δ z )(nx δ x + ny δ y + nz δ z )
= (Txx nx + Txy ny + Txz nz )δ x
+(Tyx nx + Tyy ny + Tyz nz )δ y
+(Tzx nx + Tzy ny + Tzz nz )δ z

The discussion leading to equation (4.8) is simply a statement of the relationship


between the stress tensor and the stress vector. Equation (4.8) is also known as
Cauchy’s fundamental theorem for stress. It was included here as an example to
show how the tensors as a linear combination of unit dyads are seen as operators
on vectors. A fuller development of the relationship (4.8) involves the application
of Newton’s laws of motion.2

♦♦♦

There is a special tensor, called the unit tensor defined by

δ = δxδx + δy δy + δz δz (4.9)

When this tensor operates on any vector, the result is the same vector.

4.4.1 Tensor Operations


When applying tensors, the operations of the unit dyads are simply combined linearly. More-
over, as shown in the previous example, when the vector being operated on by the tensors
are also represented as a linear combination of unit vectors, simpler formulas can result. We
summarize these operations below.

1. Tensor Addition.
X X
τ +σ = (τij + σij )δ i δ j
i=x,y,z j=x,y,z

2. Scalar Multiplication of a Tensor.


X X
ατ = (ατij )δ i δ j
i=x,y,z j=x,y,z

2
See D. C. Leigh, Nonlinear Continuum Mechanics, McGraw Hill, 1968, for a derivation of (4.8).
118 c 2006 Tomas Co, all rights reserved
°

3. Inner Product of Two Tensors.


à !
X X X
τ ·σ = τik σkj δiδj
i=x,y,z j=x,y,z k=x,y,z

4. Inner Product of A Tensor with a Vector.


à !
X X
τ ·v = τij vj δ i
i=x,y,z j=x,y,z

5. Double Dot Product.


X X
τ : σ= τij σji
i=x,y,z j=x,y,z

4.4.2 Correspondence with Matrix Operations


If the coordinate system is the orthogonal Cartesian coordinate system that is fixed in space,
then the tensor and vector operation can be represented by matrices by using the following:
     
1 0 0
δx =  0  δy =  1  δz =  0 
0 0 1

Once this vector is represented by column vectors, a close correspondence is available


between tensors and matrices. This is shown in Table 4.1

4.5 Vector Differential Operations


In this section, we explore differential operations on vectors. First, we will focus on taking
time derivative of a vector bound at a fixed point p, i.e. Vp = Vp (t). Then, we investigate
spatial differential operators of a vector field, V = V (x, y, z).

4.5.1 Vector Function of One Variable.


Consider a vector fixed at a point that is a function of a single variable, e.g. t,
V (t) = Vx (t)δ x + Vy (t)δ y + Vz (t)δ z (4.10)

We can extend the same definition used for derivatives of real functions and apply it
to vectors while using vector addition and scalar products,
c 2006 Tomas Co, all rights reserved
° 119

Table 4.1: Correspondence between Tensors and Matrices.

Vector Notations Matrix Representation

a = ax δ x + ay δ y + az δ z a = (ax , ay , az )T

a+b a + b = (ax + bx , ay + by , az + bz )T

αv α v = (αvx , αvy , αvz )T

a·b aT b

ab abT

P P
τ= i=x,y,z j=x,y,z τij δ i δ j τ = (τij )

τ +σ τ + σ = (τij + σij )

ατ α τ = (ατij )

P
τ ·σ τ σ = ( k=x,y,z τik σkj )

T
P P P
τ ·v τ v=( j=x,y,z τxj vj , j=x,y,z τyj vj , j=x,y,z τzj vj )

τ : σ Trace(τ σ)
120 c 2006 Tomas Co, all rights reserved
°

Figure 4.11: Definition of dV /dt.

dV 1
= lim (V (t + ∆t) − V (t))
dt ∆t→0 ∆t
1 ¡£ ¤
= lim Vx (t + ∆t)δ x + Vy (t + ∆t)δ y + Vz (t + ∆t)δ z
∆t→0 ∆t
£ ¤¢
− Vx (t)δ x + Vy (t)δ y + Vz (t)δ z
dVx dVy dVz
= δx + δy + δ (4.11)
dt dt dt z

To illustrate the definition geometrically, (4.11) is shown in Figure 4.11. Note that the
vector dV /dt does not necessarily point in the same direction as V .
Applying (4.11) to different operations, the following identities follow immediately:

1. Derivative of vector sum.

d dU dV
(U (t) + V (t)) = + (4.12)
dt dt dt
2. Derivative of scalar products.
d dφ dV
(φ(t)V (t)) = V +φ (4.13)
dt dt dt

3. Derivative of dot products.


d dU dV
(U (t) · V (t)) = ·V +U · (4.14)
dt dt dt

4. Derivative of cross products.


d dU dV
(U (t) × V (t)) = ×V +U × (4.15)
dt dt dt
c 2006 Tomas Co, all rights reserved
° 121

4.5.2 Differential operators on Vector Fields

In this section, we will consider both a scalar field, φ, and a vector field V in which both fields
are defined as function of its position in a coordinate system, e.g. φ(x, y, z) and V (x, y, z),
respectively.

4.5.2.1 Gradients

Consider a three dimensional scalar field φ(x, y, z). An example of a scalar field is the
temperature distribution in the three dimension space, T (x, y, z). Usually the gradation of
scalar fields are used to determine the driving forces during transport of mass, momentum or
energy. Thus, it is important to generate a flow field based on the gradations to determine
the direction and magnitude of the driving forces. The flow field, or vector field, that is
generated by the gradation of a scalar field is called the gradient vector field. At each
point (x, y, z), there corresponds a vector attached to the point given by

∂φ ∂φ ∂φ
gradient(φ(x, y, z)) = δx + δy + δ (4.16)
∂x ∂y ∂z z

This definition is further simplified if we introduce a differential operator called the gradient
operator, or grad for short, denoted by ∇ and defined by

∂ ∂ ∂
∇ = δx + δy + δz (4.17)
∂x ∂y ∂z

Thus,
gradient(φ(x, y, z)) = ∇φ(x, y, z)

To graphically illustrate the definition of gradient, consider the special case in which
the scalar field is defined only on the x–y plane. As shown in Figure 4.12, we need to take
the partial derivatives with respect to each dimension. Each of these partial derivatives
determine the magnitude of the x or y components which are then added vectorially to yield
the gradient vector at that point.
The gradient vector is also sometimes used to determine the rate of change in φ along
a path s in the domain. At any point, the differential of φ(x, y, z) is given by

∂φ ∂φ ∂φ
dφ(x, y, z) = dx + dy + dz
∂x ∂y ∂z

Along the path s, the rate of change in φ is obtained by dividing by ds,

dφ(x, y, z) ∂φ dx ∂φ dy ∂φ dz
= + + (4.18)
ds ∂x ds ∂y ds ∂z ds
122 c 2006 Tomas Co, all rights reserved
°

Figure 4.12: Evaluation of ∇φ(x, y).

The right hand side of the equation can now be factored into a dot product,
µ ¶
dφ(x, y, z) dx dy dz
= (∇φ) · δ + δ + δ
ds ds x ds y ds z
= (∇φ) · t (4.19)

where
dx dy dz
t= δx + δy + δz
ds ds ds
is a unit vector tangent to the path s at the particular point (x, y, z) in the path as shown
in Figure 4.13. The term dφ/ds is known as the directional derivative along the direction
t.
By considering (4.19) and the definition of a dot product,

= k∇φk cos θ (4.20)
ds
where θ is the angle between ∇φ and t. Thus, if the vector ∇φ is pointed at the same
direction as t, i.e. θ = 0 and cos θ = 1, we obtain the maximum rate of increase in the scalar
field at that particular point. In other words, at a particular point (xo , yo , zo ), the vector
∇φ at this point indicates both the maximum rate of increase of φ within its differential
neighborhood and the direction under which this maximum is achieved.
Another consequence of (4.20) is that when θ = 90o , dφ/ds = 0, i.e. φ(x, y, z) is
constant in the direction perpendicular to the gradient vector, ∇φ. To illustrate, if φ =
φ(x, y), a contour map such as shown in Figure 4.14 also indicates partial gradient field
c 2006 Tomas Co, all rights reserved
° 123

Figure 4.13: The unit tangent vector along path s.

information. For instance, the gradient vectors at contour lines are directed normal to the
contour curves. Also, if the distance between two successive contour levels is farther at a
point, the gradient vector has a smaller magnitude at that point.

4.5.2.2 Divergence

Consider now a vector field V , where attached at a point (x, y, z) is a vector


V (x, y, z) = Vx (x, y, z)δ x + Vy (x, y, z)δ y + Vz (x, y, z)δ z
Figure 4.15 shows a vector field for a two dimensional domain. The vectors at different
points will generally have different magnitudes and direction.
We will assume that the vector field is continuous, i.e. that the vector components
Vx (x, y, z), Vy (x, y, z) and Vz (x, y, z) have finite first order partial derivatives. The diver-
gence of a vector field at a point is then given as the sum of partial derivatives of the
components along their respective axes,
∂Vx ∂Vy ∂Vz
divergence(V (x, y, z)) = + + (4.21)
∂x ∂y ∂z
In terms of the ∇ operator, the divergence is written as ∇ · V ,
µ ¶
∂ ∂ ∂ ¡ ¢
∇·V = δx + δy + δz · Vx δ x + Vy δ y + Vz δ z
∂x ∂y ∂z
124 c 2006 Tomas Co, all rights reserved
°

Figure 4.14: Contour maps for a two dimensional scalar field. Note that k∇φk at Q is greater
than k∇φk at P .

Figure 4.15: A vector field in two dimensional domain.


c 2006 Tomas Co, all rights reserved
° 125

Figure 4.16: Evaluation of the divergence of a vector field at a point.

∂Vx ∂Vy ∂Vz


= δx · δx + δx · δy + δx · δ
∂x ∂x ∂x z
∂Vx ∂Vy ∂Vz
+δ y · δx + δy · δy + δy · δ
∂y ∂y ∂y z
∂Vx ∂Vy ∂Vz
+δ z · δx + δz · δy + δz · δ
∂z ∂z ∂z z
∂Vx ∂Vy ∂Vz
= + +
∂x ∂y ∂z
Based on this definition, the evaluation of divergence is shown in Figure 4.16. Note
that the divergence of a vector field will yield a scalar field. This is in direct to contrast to
gradients, for which the gradient of a scalar field will result in a vector field.
In physical terms, if the vector field is specifically a velocity field, the divergence at
the point of a velocity field is a measure of how much a property such as concentration or
mass diverges at that point due to the changes in velocities along each axis. In fact, the
continuity equation (or differential mass balance) is given by
∇ · (ρ(x, y, z)V (x, y, z)) = 0
where ρ is the density and V is the velocity field, both of which are dependent on the position
(x, y, z). If ∇ · (ρV ) > 0 at some point P , then we say that P is a source point. On the
other hand, if ∇ · (ρV ) < 0 at point Q, then Q is said to be a sink point.
126 c 2006 Tomas Co, all rights reserved
°

Figure 4.17: Evaluation of the curl of a vector field at a point.

4.5.2.3 Curl

Due to the distribution in a vector field, neighboring vectors will affect the differential volume
at a point to twirl or rotate. Figure 4.17 shows the effects of vectors that lie in the x–y plane
at a fixed z level. Next we define Cz δ z to be the vector to describe its tendency to rotate
based on the rate of increase of the y-component along the x-axis and the rate of increase of
the x-components along the y-axis. Using the right-hand screw convention, the total effect
is given by
µ ¶
Vy (x + ∆x, y, z) − Vy (x, y, z) Vx (x, y + ∆y, z) − Vy (x, y, z)
Cz = lim −
∆→0 ∆x ∆y
∂Vy ∂Vx
= −
∂x ∂y

Following the same analysis in the x–z plane, we get Cy δ y

∂Vx ∂Vz
Cy = −
∂z ∂x
and in the y–z plane, we have Cx δ x , where
∂Vz ∂Vy
Cx = −
∂y ∂z
c 2006 Tomas Co, all rights reserved
° 127

Adding the three vectors yields a vector known as the Curl of the vector field at the point
(x, y, z),
µ ¶ µ ¶ µ ¶
∂Vz ∂Vy ∂Vx ∂Vz ∂Vy ∂Vx
Curl(V (x, y, z)) = − δx + − δy + − δz (4.22)
∂y ∂z ∂z ∂x ∂x ∂y
Using the ∇ operator, the curl is written as the cross product of ∇ with V ,

µ ¶
∂ ∂ ∂ ¡ ¢
∇×V = δx + δy + δz × Vx δ x + Vy δ y + Vz δ z
∂x ∂y ∂z
∂Vx ∂Vy ∂Vz
= δx × δx + δx × δy + δx × δ
∂x ∂x ∂x z
∂Vx ∂Vy ∂Vz
+δ y × δx + δy × δy + δy × δ
∂y ∂y ∂y z
∂Vx ∂Vy ∂Vz
+δ z × δx + δz × δy + δz × δz
µ ∂z ¶ µ∂z ¶ ∂z µ ¶
∂Vz ∂Vy ∂Vx ∂Vz ∂Vy ∂Vx
= − δx + − δy + − δz
∂y ∂z ∂z ∂x ∂x ∂y
= Curl(V )

Since under the cartesian coordinates the unit vectors have constant directions, the
following mnemonic will still work:
 
δx δy δz
 
 ∂ ∂ ∂ 
∇ × V = det  ∂x ∂y ∂z 


 
Vx Vy Vz
However, it is crucial to note that the determinant formula will not yield the correct curl for-
mula when using cylindrical coordinate systems or spherical coordinate systems, because at
both these coordinate systems the unit vectors are pointing at different directions depending
on the position in the vector field. (These coordinate systems will be discussed further in
section 4.6 and 4.7).
When the vector field V is specifically a velocity field, ∇ × V is also known as the
vorticity. If a solid particle of differential volume is positioned at a point P in the velocity
field, then that particle will experience an angular velocity Ω whose direction (based on the
right-hand screw rule) will point in the same direction as ∇ × V . However, as one can see
from Figure 4.17, the z-component of the angular velocity Ωz is the average of the two effects
in Cz , i.e. Ωz = (1/2)Cz . This means that to obtain the magnitude of the angular velocity,
we just need to divide the curl by 2, thus
1
Ω(x, y, z) = ∇ × V
2
128 c 2006 Tomas Co, all rights reserved
°

4.5.2.4 Laplacian

Taking the gradient of a scalar field, φ(x, y, z) will result in a vector field, ∇φ. After taking
the divergence of this resulting vector field, we obtain the operation known as the Laplacian,

Laplacian(φ(x, y, z)) = ∇ · (∇φ) (4.23)


∂2φ ∂ 2φ ∂ 2φ
= + + 2 (4.24)
∂x2 ∂y 2 ∂z

In some books, the Laplacian notation for φ is further simplified as ∇2 φ.


As a simple physical interpretation, consider the heat conduction process in a three
dimensional solid. The energy balance is

Rate of Change in Energy = Accumulation or Negative Divergence of Energy Flow

With constant heat conductivity, density and heat capacity, the divergence of energy flow
can be replaced by a proportionality to the divergence of the temperature gradient. Thus,
∂T
= α∇2 T
∂t
where α = k/(ρCp), k is thermal conductivity, ρ is density and Cp is the heat capacity.

4.5.2.5 Vector Differential Identities

Let φ(x, y, z) be a scalar field, V and U are vector fields. We list below some of the identities
that can be easily obtained by applying the definition directly to both sides of the equations.

∇ · φV = φ∇ · V + V · ∇φ (4.25)
∇ × φV = φ∇ × V + (∇φ) × V (4.26)
∇ · (U × V ) = V ·∇×U − U ·∇×V (4.27)
∇ × (U × V ) = V · ∇U − U · ∇V + U ∇ · V − V ∇ · U (4.28)
∇(U · V ) = U · ∇V + V · ∇U + U × (∇ × V ) + V × (∇ × U ) (4.29)
∇ × ∇φ = 0 (4.30)
∇·∇×V = 0 (4.31)
∇ × (∇ × V ) = ∇(∇ · V ) − ∇2 V (4.32)
c 2006 Tomas Co, all rights reserved
° 129

Example 4.2 . Proof of (4.29)

µ ¶
∂ux ∂uy ∂uz ∂vx ∂vy ∂vz
∇(U · V ) = δ x vx + vy + vz + ux + uy + uz
∂x ∂x ∂x ∂x ∂x ∂x
µ ¶
∂ux ∂uy ∂uz ∂vx ∂vy ∂vz
+δ y vx + vy + vz + ux + uy + uz
∂y ∂y ∂y ∂y ∂y ∂y
µ ¶
∂ux ∂uy ∂uz ∂vx ∂vy ∂vz
+δ z vx + vy + vz + ux + uy + uz
∂z ∂z ∂z ∂z ∂z ∂z
(4.33)

We can expand U × (∇ × V ) to be
µ ¶
∂vy ∂vx ∂vx ∂vz
U × (∇ × V ) = δ x uy − uy − uz + uz
∂x ∂y ∂z ∂x
µ ¶
∂vz ∂vy ∂vy ∂vx
+δ y uz − uz − ux + ux
∂y ∂z ∂x ∂y
µ ¶
∂vx ∂vz ∂vz ∂vy
+δ z ux − ux − uy + uy (4.34)
∂z ∂x ∂y ∂z

and combined with (U · ∇) V ,


µ ¶
∂ ∂ ∂ ¡ ¢
(U · ∇) V = ux + uy + uz vx δ x + vy δ y + vz δ z
∂x ∂y ∂z
µ ¶
∂vx ∂vx ∂vx
= δ x ux + uy + uz
∂x ∂y ∂z
µ ¶
∂vy ∂vy ∂vy
+δ y ux + uy + uz
∂x ∂y ∂z
µ ¶
∂vz ∂vz ∂vz
+δ z ux + uy + uz (4.35)
∂x ∂y ∂z

to yield,
µ ¶
∂vx ∂vy ∂vz
U × (∇ × V ) + (U · ∇) V = δ x ux + uy + uz
∂x ∂x ∂x
µ ¶
∂vx ∂vy ∂vz
+δ y ux + uy + uz
∂y ∂y ∂y
µ ¶
∂vx ∂vy ∂vz
+δ z ux + uy + uz (4.36)
∂z ∂z ∂z
130 c 2006 Tomas Co, all rights reserved
°

By interchanging the role of U and V ,


µ ¶
∂ux ∂uy ∂uz
V × (∇ × U ) + (V · ∇) U = δ x vx + vy + vz
∂x ∂x ∂x
µ ¶
∂ux ∂uy ∂uz
+δ y vx + vy + vz
∂y ∂y ∂y
µ ¶
∂ux ∂uy ∂uz
+δ z vx + vy + vz (4.37)
∂z ∂z ∂z
By adding the left hand sides of (4.36) and (4.37) we get the left hand side of
(4.33). Thus we prove the identity given in (4.29).

♦♦♦

4.6 Cylindrical Coordinate System


The cylindrical coordinate system is a coordinate system which can be derived from the
rectangular coordinate system using the following nonlinear transformations:
p
r = x2 + y 2 (4.38)
−1
θ = tan (y/x) (4.39)
z = z (4.40)
This transformation is invertible, i.e. the rectangular coordinate can be recovered when
given the cylindrical coordinates,
x = r cos θ (4.41)
y = r sin θ (4.42)
z = z (4.43)
The cylindrical coordinate system is shown in Figure 4.18
A summary of the relationships between cylindrical and rectangular coordinates is
given in Table 4.2. Although these relationships can be obtained by mathematical manipu-
lations, it is still instructive to see the geometric implications.
At a point (r, θ, z), the pair of unit vectors δr and δ θ is just the pair of unit vectors δ x
and δ y rotated counter clockwise by an angle θ. By defining a rotation operator,3
 
cos θ sin θ 0
Rr→c =  − sin θ cos θ 0  (4.44)
0 0 1
3
Note that the rotation operator in (4.44) would rotate a vector clockwise by an angle θ. However, since
we are rotating the reference axis, this operator would do the reverse, i.e. rotate the axes counterclockwise.
c 2006 Tomas Co, all rights reserved
° 131

Table 4.2: Relationship between Rectangular and Cylindrical Coordinates

Rectangular Cylindrical

∂ ∂ ∂ ∂ ∂
− ¡ sinr θ ¢ ∂θ
¡ ¢ ∂
∂x
= cos θ ∂r ∂r
= cos θ ∂x + sin θ ∂y
∂ ∂
∂y
= sin θ ∂r + cosr θ ∂θ ∂ ∂
∂θ

= −r sin θ ∂x ∂
+ r cos θ ∂y
∂ ∂ ∂ ∂
∂z
= ∂z ∂z
= ∂z

δ x = cos θδ r − sin θδ θ δ r = cos θδ x + sin θδ y


δ y = sin θδ r + cos θδ θ δ θ = − sin θδ x + cos θδ y
δz = δz δz = δz

V = Vx δ x + Vy δ y + Vz δ z V = Vr δ r + Vθ δ θ + Vz δ z

Vx = Vr cos θ − Vθ sin θ Vr = Vx cos θ + Vy sin θ


Vy = Vr sin θ + Vθ cos θ Vθ = −Vx sin θ + Vy cos θ
Vz = Vz Vz = Vz

∂δ r /∂θ = δ θ
∂δ i /∂j = 0
∂δ θ /∂θ = −δ r
for i, j = x, y, z
zero for all other cases

∂ ∂ ∂ ∂
∇ = δ x ∂x + δ y ∂y + δ z ∂z ∇ = δ r ∂r + δ θ 1r ∂
∂θ

+ δ z ∂z
132 c 2006 Tomas Co, all rights reserved
°

Figure 4.18: Cylindrical coordinate system.

we have ¢t
(δ r , δ θ , δ z )t = Rr→c δ x , δ y , δ z
¡
(4.45)
And since Rr→c is orthogonal,
¢t t
(δ r , δ θ , δ z )t
¡
δx, δy , δz = Rr→c (4.46)

Applying (4.45) and (4.46) to a vector in rectangular coordinates,


V = Vx δ x + Vy δ y + Vz δ z
= (Vx , Vy , Vz )(δ x , δ y , δ z )t
t
= (Vx , Vy , Vz )Rr→c (δ r , δ θ , δz)t (4.47)
Thus,
t
(Vr , Vθ , Vz ) = (Vx , Vy , Vz )Rr→c (4.48)
or
(Vr , Vθ , Vz )t = Rr→c (Vx , Vy , Vz )t (4.49)

To obtain the relationship between the partial differential operators in each represen-
tation, the chain rule yields
 ∂   ∂x ∂y ∂z   ∂ 
∂r ∂r ∂r ∂r ∂x
    
∂y ∂
 ∂
  ∂x ∂z
 
  =   
 ∂θ   ∂θ ∂θ ∂θ  ∂y 
    
∂ ∂x ∂y ∂z ∂
∂z ∂z ∂z ∂z ∂z
c 2006 Tomas Co, all rights reserved
° 133

 ∂ 
  ∂x
cos θ sin θ 0  

 
=  −r sin θ r cos θ 0  
 ∂y


0 0 1  

∂z
 ∂ 
∂x
 

 
= Dr→c Rr→c 
 ∂y

 (4.50)
 

∂z

where, Dr→c = diag(1, r, 1).


To derive the ∇ operator in cylindrical coordinates.
µ ¶t
∂ ∂ ∂
∇ = (δ x , δ y , δ z ) , ,
∂x ∂y ∂z
" µ ¶t #
t −1 ∂ ∂ ∂
= [(δ r , δ θ , δ z )Rr→c ] Rr→c Dr→c , ,
∂r ∂θ ∂z
µ ¶t
1 ∂ ∂ ∂
= (δ r , δ , δ ) , , (4.51)
r θ z ∂r ∂θ ∂z

Since the unit vectors δ r and δ θ are dependent on θ, the gradient formulas such as
curls, divergences and Laplacian are more complicated.
The partial derivative of δ r with respect to θ can be interpreted geometrically using
the definition
∂δ r δ (r, θ + ∆θ) − δ r (r, θ)
= lim r (4.52)
∂θ ∆θ→0 ∆θ
Since (4.52) requires the subtraction of two vectors positioned at different points, we
need to treat the unit vectors as free vectors. As shown in the Figure 4.19, the limit yields
a vector that points in the same direction as δ θ . The norm of this limit vector is
¯ ¯
¯ lim δ r (r, θ + ∆θ) − δ r (r, θ) ¯ = lim 2|δ r | sin ∆θ/2 = 1
¯ ¯
¯∆θ→0 ∆θ ¯ ∆θ→0 ∆θ
thus,
∂δ r
= δθ (4.53)
∂θ
Similarly, for ∂δ θ /∂θ,
∂δ θ δ (r, θ + ∆θ) − δ θ (r, θ)
= lim θ (4.54)
∂θ ∆θ→0 ∆θ
134 c 2006 Tomas Co, all rights reserved
°

Figure 4.19: Unit vectors along r at different θ positions.

The length of this limit vector is


¯ ¯
¯ lim θ
¯ δ (r, θ + ∆θ) − δ θ (r, θ) ¯ = lim 2|δ θ | sin ∆θ/2 = 1
¯
¯∆θ→0 ∆θ ¯ ∆θ→0 ∆θ
As shown in Figure 4.20, the limit yields a vector that points in opposite direction of δ r .

Figure 4.20: Unit vectors along θ at different θ positions.

Thus,
∂δ θ
= − δr (4.55)
∂θ
Due to identities (4.53) and (4.55), the gradients, curls and Laplacian in cylindrical
coordinates need to be performed using the basic definitions.
For example, the divergence formula in cylindrical coordinates is given by
µ ¶
∂ 1 ∂ ∂
∇·v = δr + δ + δz · (Vr δ r + Vθ δ θ + Vz δ z )
∂r r θ ∂θ ∂z
µ ¶
∂Vr ∂Vθ ∂Vz
= δr · δ + δ + δ
∂r r ∂r θ ∂r z
c 2006 Tomas Co, all rights reserved
° 135

µ ¶
1 ∂Vr ∂Vθ ∂Vz
+ δθ · δ + δ + δ
r ∂θ r ∂θ θ ∂θ z
µ ¶
1 ∂δ r ∂δ θ
+ δ θ · Vr + Vθ
r ∂θ ∂θ
µ ¶
∂Vr ∂Vθ ∂Vz
+δ z · δ + δ + δ
∂z r ∂z θ ∂z z
∂Vr 1 ∂Vθ Vr ∂Vz
= + + + (4.56)
∂r r ∂θ r ∂z

For the curl,


µ ¶
∂ 1 ∂ ∂
∇×v = δr + δ + δz × (Vr δ r + Vθ δ θ + Vz δ z )
∂r r θ ∂θ ∂z
µ ¶
∂Vr ∂Vθ ∂Vz
= δr × δ + δ + δ
∂r r ∂r θ ∂r z
µ ¶
1 ∂Vr ∂Vθ ∂Vz
+ δθ × δ + δ + δ
r ∂θ r ∂θ θ ∂θ z
µ ¶
1 ∂δ r ∂δ θ
+ δ θ × Vr + Vθ
r ∂θ ∂θ
µ ¶
∂Vr ∂Vθ ∂Vz
+δ z × δ + δ + δ
∂z r ∂z θ ∂z z
∂Vθ ∂Vz
= δz − δ
∂r µ ∂r θ ¶
1 ∂Vr ∂Vz
+ − δ + δ
r ∂θ z ∂θ r
1
+ Vθ δ z
r
∂Vr ∂Vθ
+ δθ − δr
µ ∂z ∂z ¶ µ ¶ µ µ ¶¶
1 ∂Vz ∂Vθ ∂Vz ∂Vr ∂Vθ 1 ∂Vr
= − δr + − δ + δ δ + − + Vθ δz
r ∂θ ∂z ∂r θ ∂z θ θ ∂r r ∂θ
µ ¶ µ ¶ µ ¶
1 ∂Vz ∂Vθ ∂Vz ∂Vr 1 ∂ 1 ∂Vr
= − δr + − δ + δ δ + (rVθ ) − δz
r ∂θ ∂z ∂r θ ∂z θ θ r ∂r r ∂θ
(4.57)

A summary of important differential operations is conveniently listed in Bird, Stewart


and Lightfoot, “Transport Phenomena”, J.Wiley, 1960, pp. 739B.
136 c 2006 Tomas Co, all rights reserved
°

4.7 Spherical Coordinate System

The spherical coordinate system is a coordinate system which can be derived from the
rectangular coordinate system using the following nonlinear transformation:

p
r = x2 + y 2 + z 2 (4.58)
p
θ = tan−1 ( x2 + y 2 /z) (4.59)
φ = tan−1 (y/x) (4.60)

whose inverse transformation is given by

x = r sin θ cos φ (4.61)


y = r sin θ sin φ (4.62)
z = r cos θ (4.63)

A diagram showing the relationship between the rectangular and spherical coordinate system
is shown in the following figure: A summary of important transformation between the two

Figure 4.21: The spherical coordinate system.


c 2006 Tomas Co, all rights reserved
° 137

coordinate systems is given below:

Rectangular Spherical

∂ ∂
¡ cos θ cos φ ¢ ∂ ∂ ∂ ∂
∂x
= sin θ cos φ ∂r
+ r ∂θ ∂r
= sin θ cos φ ∂x + sin θ sin φ ∂y
− rsin
¡ φ¢ ∂

sin θ ∂φ ¡ + cos θ ∂z
∂ ∂
+ cos θrsin φ ∂θ ∂ ∂ ∂
¢ ∂
∂y
= sin θ sin φ ∂r ∂θ
= r cos θ cos φ ∂x + r cos θ sin φ ∂y

+ rcos
¡ φ¢ ∂
sin θ ¡∂φ ¢
− r sin θ ∂z
∂ ∂ ∂
∂ ∂
= cos θ ∂r − sinr θ ∂θ∂
∂φ
= −r sin θ sin φ ∂x + r sin θ cos φ ∂y
∂z

δ x = sin θ cos φδ r + cos θ cos φδ θ δ r = sin θ cos φδ x + sin θ sin φδ y


− sin φδ φ + cos θδ z
δ y = sin θ sin φδ r + cos θ sin φδ θ δ θ = cos θ cos φδ x + cos θ sin φδ y
+ cos φδ φ − sin θδ z
δ z = cos θδ r − sin θδ θ δ φ = − sin φδ x + cos φδ y

V = Vx δ x + Vy δ y + Vz δ z V = Vr δ r + Vθ δ θ + Vφ δ φ

Vx = sin θ cos φVr + cos θ cos φVθ Vr = sin θ cos φVx + sin θ sin φVy
− sin φVφ + cos θVz
Vy = sin θ sin φVr + cos θ sin φVθ Vθ = cos θ cos φVx + cos θ sin φVy
+ cos φVφ − sin θVz
Vz = cos θVr − sin θVθ Vφ = − sin φVx + cos φVy
∂δ r /∂θ = δ θ
∂δ θ /∂θ = −δ r
∂δ i /∂j = 0 ∂δ r /∂φ = δ φ sin θ
for i, j = x, y, z ∂δ θ /∂φ = δ φ cos θ
∂δ φ /∂φ = −δ r sin θ − δ θ cos θ
zero for all other cases

∂ ∂ ∂ ∂
∇ = δ x ∂x + δ y ∂y + δ z ∂z ∇ = δ r ∂r + δ θ 1r ∂
∂θ
1
+ δ φ r sin ∂
θ ∂φ

The geometric relationship among the unit vectors in rectangular and spherical coordinate
systems involve two consecutive rotations. At a point (r, θ, φ), the unit vectors are first
rotated φ radians counterclockwise along δ x , δ y plane which can be achieved by
 
cos φ sin φ 0
Rrs1 =  − sin φ cos φ 0 
0 0 1
138 c 2006 Tomas Co, all rights reserved
°

Then another rotation of θ radians clockwise is applied on the three unit vectors along the
transformed δ x , δ z plane using Rrs2 .
 
cos θ 0 − sin θ
Rrs2 =  0 1 0 
sin θ 0 cos φ

Finally, the vectors that were originally δ x , δ y and δ z are now renamed δ θ , δ φ and δ r ,
respectively. The relationship is thus given by

(δ θ , δ φ , δ r )t = Rrs2 Rrs1 (δ x , δ y , δ z )t

Since the usual convention uses the specific order of (r, θ, φ), we could add another reordering
operator  
0 0 1
Ereord =  1 0 0 
0 1 0
That is,
(δ r , δ θ , δ φ )t = Ereord Rrs2 Rrs1 (δ x , δ y , δ z )t = Rr→s (δ x , δ y , δ z )t
where  
sin θ cos φ sin θ sin φ cos θ
Rr→s =  cos θ cos φ cos θ sin φ − sin θ  (4.64)
− sin φ cos φ 0
Since Rr→s is orthogonal, the inverse relationship is given by

(δ x , δ y , δ z )t = Rrs
t
(δ r , δ θ , δ φ )t

A vector in rectangular coordinates is given by

V = (Vx , Vy , Vz )(δ x , δ y , δ z )t
t
= (Vx , Vy , Vz )Rr→s (δ r , δ θ , δ φ )t

hence,
(Vr , Vθ , Vφ )t = Rr→s (Vx , Vy , Vz )t (4.65)
The partial differential operators between the rectangular and spherical coordinate system
is again obtained by using the chain rule,
 ∂   ∂x ∂y ∂z  ∂ 
∂r ∂r ∂r ∂r ∂x
    
∂ ∂x ∂y ∂z ∂
    
  =   
 ∂θ   ∂θ ∂θ ∂θ  ∂y 
    
∂ ∂x ∂y ∂z ∂
∂φ ∂φ ∂φ ∂φ ∂z
c 2006 Tomas Co, all rights reserved
° 139

  ∂ 
sin θ cos φ sin θ sin φ cos θ ∂x
  

  
 r cos θ cos φ r cos θ sin φ −r sin θ
=   
 ∂y 
  
−r sin θ sin φ r sin θ cos φ 0 ∂
∂z
 ∂ 
∂x
 

 
= Dr→s Rrs 
 ∂y

 (4.66)
 

∂z

where, Dr→s = diag(1, r, r sin θ).


The ∇ operator in spherical coordinates can then be obtained as follows,
µ ¶t
∂ ∂ ∂
∇ = (δ x , δ y , δ z ) , ,
∂x ∂y ∂z
" µ ¶t #
£ ¤ t −1 ∂ ∂ ∂
= (δ r , δ θ , δ φ )Rr→s Rr→s Dr→s , ,
∂r ∂θ ∂φ
µ ¶t
1 1 ∂ ∂ ∂
= (δ r , δ , δ ) , , (4.67)
r θ r sin θ z ∂r ∂θ ∂φ

Just like the cylindrical coordinate systems, the unit vectors in spherical coordinate
systems are not fixed in space. Based on Figure 4.22,
∂δ r
= δθ (4.68)
∂θ

Figure 4.22: Unit vectors along r at different θ positions.

While, based on Figure 4.23,


∂δ θ
= − δr (4.69)
∂θ
140 c 2006 Tomas Co, all rights reserved
°

Figure 4.23: Unit vectors along θ at different θ positions.

For the differential changes in φ, the rotation is obtained on a projection on a horizontal


plane. Based on Figure 4.24, ∂δ φ /∂φ will yield a unit vector that is directed inward along
the horizontal plane. From Figure 4.25, we see that this vector is a vector combination of δ r
and δ θ , resulting in

∂δ φ
= − cos δ θ − sin θδ r (4.70)
∂φ

Figure 4.24: Unit vectors along φ at different φ positions projected on the horizontal plane.

Figure 4.25: The horizontal inward unit vector as sum of vectors in the r and θ directions.

By projecting δ r onto the horizontal plane, we have the vectors shown in Figure 4.26
c 2006 Tomas Co, all rights reserved
° 141

Figure 4.26: Changes in δ r projected onto the horizontal plane.

Then the result becomes,


∂δ r
= sin θδ φ (4.71)
∂φ

Similarly, by projecting δ r onto the horizontal plane,

∂δ θ
= cos θδ φ (4.72)
∂φ

To obtain the divergence, we need to distribute the operations.


µ ¶
∂ 1 ∂ 1 ∂ ¡ ¢
∇·V = δr + δθ + δφ · Vr δ r + Vθ δ θ + Vφ δ φ
∂r r ∂θ r sin θ ∂φ
µ ¶
∂Vr ∂Vθ ∂Vφ
= δr · δ + δ + δ
∂r r ∂r θ ∂r φ
µ ¶
1 ∂Vr ∂Vθ ∂Vφ
+ δθ · δ + δ + δ
r ∂θ r ∂θ θ ∂θ φ
µ ¶
1 ∂δ r ∂δ θ
+ δ θ · Vr + Vθ
r ∂θ ∂θ
µ ¶
1 ∂Vr ∂Vθ ∂Vφ
+ δ · δ + δ + δ
r sin θ φ ∂φ r ∂φ θ ∂φ φ
∂δ φ
µ ¶
1 ∂δ r ∂δ θ
+ δ · Vr + Vθ + Vφ
r sin θ φ ∂φ ∂φ ∂φ
∂Vr 1 ∂Vθ Vr 1 ∂Vφ Vθ cos θ
= + +2 + +
∂r r ∂θ r r sin θ ∂φ r sin θ
2
1 ∂(r Vr ) 1 ∂(Vθ sin θ) 1 ∂Vφ
= 2
+ + (4.73)
r ∂r r sin θ ∂θ r sin θ ∂φ

For the curl,


142 c 2006 Tomas Co, all rights reserved
°

µ ¶
∂ 1 ∂ 1 ∂ ¡ ¢
∇×V = δr + δθ + δφ × Vr δ r + Vθ δ θ + Vφ δ φ
∂r r ∂θ r sin θ ∂φ
µ ¶
∂Vr ∂Vθ ∂Vφ
= δr × δ + δ + δ
∂r r ∂r θ ∂r φ
µ ¶
1 ∂Vr ∂Vθ ∂Vφ
+ δθ × δ + δ + δ
r ∂θ r ∂θ θ ∂θ φ
µ ¶
1 ∂δ r ∂δ θ
+ δ θ × Vr + Vθ
r ∂θ ∂θ
µ ¶
1 ∂Vr ∂Vθ ∂Vφ
+ δ × δ + δ + δ
r sin θ φ ∂φ r ∂φ θ ∂φ φ
∂δ φ
µ ¶
1 ∂δ r ∂δ θ
+ δ × Vr + Vθ + Vφ
r sin θ φ ∂φ ∂φ ∂φ
∂Vθ ∂Vφ 1 ∂Vr 1 ∂Vφ Vθ
= δφ − δθ − δφ + δr + δφ
∂r ∂r r ∂θ r ∂θ r
1 ∂Vr 1 ∂Vθ Vφ Vφ cos θ
+ δθ − δr − δθ + δ
r sin θ ∂φ r sin θ ∂φ r r sin θ r
µ ¶ µ ¶
1 ∂(Vφ sin θ) 1 ∂Vθ 1 ∂(rVφ ) 1 ∂Vr
= + − δr + − + δθ
r sin θ ∂θ r sin θ ∂φ r ∂r r sin θ ∂φ
µ ¶
1 ∂(rVθ ) 1 ∂Vr
+ − δφ (4.74)
r ∂r r ∂θ

A summary of other important differential operations is listed in Bird, Stewart and


Lightfoot, “Transport Phenomena”, J.Wiley, 1960, pp. 739C-D.
Chapter 5

Integral Theorems

In this chapter we discuss different integral operations such as line integral, surface integral
and volume integral. Based on these integrals we have important theorems such as Green’s
Lemma, the Divergence Theorem and Stoke’s Theorem that are useful in several applications
in vector analysis.

5.1 Line Integrals


Line integrals are specified by three components:

1. The path of integration, C(x, y, z), which is a continuous curve in (x, y, z).

2. The integrand, F (x, y, z), which is a scalar function.

3. The differential, which could either be dx, dy, dz or ds where


p
ds = dx2 + dy 2 + dz 2 (5.1)

Based on these three components, the line integrals are defined as the limit of a sum-
mation of the products of the integrand F (x, y, z) and either of the subintervals, ∆x, ∆y,
∆z or ∆s, along the path of integration.
Z n
X
F (x, y, z)dx = lim F (xi , yi , zi )∆xi (5.2)
C ∆xi →0,n→∞
i=1
( line integral with respect to x
along the path C )

143
144 c 2006 Tomas Co, all rights reserved
°

Z n
X
F (x, y, z)ds = lim F (xi , yi , zi )∆si (5.3)
C ∆si →0,n→∞
i=1
( line integral with respect to the arc traversed, s,
along the path C )
R R
Similar definitions apply to c
F (x, y, z)dy, and c
F (x, y, z)dz.
Suppose the integrand F and integration path C are functions only of x and y, Figure
5.1 illustrates the area
R interpretation of the line integrals. The line integral with respect to
the arc of the path, C F (x,Ry)ds is the area under the curve C. On the other hand, the line
integral with respect to x, C F (x, y)dx is the area projected onto the plane formed by the
x-axis and the F -axis.

Figure 5.1: Area interpretation of line integrals.

Of course, this area interpretation applies only to segments of the integration path
where
R C(x, y) is single-valued with respect to x. This means that when evaluating
C
F (x, y)dx, the integration path C will have to be partitioned such that within each
segment the subpath is single-valued with respect to x. For instance, in Figure 5.2, the
integration path from A to B will have to be partitioned into segment ADE, segment EF
and segment F GB.
Thus, for the integration path shown in Figure 5.2, the line integral with respect to x
is given by the sum of three line integrals, i.e.
Z Z Z Z
F (x, y)dx = F (x, y)dx + F (x, y)dx + F (x, y)dx (5.4)
C [ADE] [EF ] [F GB]
c 2006 Tomas Co, all rights reserved
° 145

Figure 5.2: A curve in which the projection of C onto x or y is not single valued.

This area interpretation is more difficult to see when the curve is three dimensional,
i.e. when C = C(x, y, z). One possible visualization is to imagine a creature mining through
the ground along a path C for substance A. The ground has different concentration distri-
bution of A. Let F (x, y, z) be the amount of A gathered per unit length traveled. Then,
as the creature travels along a differential path, ds, it accumulates an amount F (x, y, z)ds.
The
R total amount of A gathered along the whole R path C is then given by the line integral
C
F (x, y, z)ds. On the other hand, the integral C
F (x, y, z)dx is the total amount gathered
only along the x-direction traveled.
R
For the scenario of the mining creature given above, the line integral C F (x, y, z)dx
does not appear to be as relevant as the line integral with respect to s. However, these line
integrals will be quite important when we use line integrals to define and compute surface
integrals or volume integrals. For surface integrals, the differential surface will be often be
described by the products such as dxdy, dxdz or dydz. While for volume integrals, the
differential volume will often be described by the product dxdydz.

5.1.1 The Path of Integration

The path of integration has a sectionally smooth and continuous curve that can either be
open or closed. For a closed path, the starting point of the path coincides with the end point
of the path. In either case, the direction of the path is crucial during integration.
There are several ways of describing this path. One method is to parameterize the
changing position of the curve by a real parameter, say t, starting with t = 0 and ending
with t = 1.1 Another method is to use explicit relationships. For instance, if the curve reside
in the (x, y) plane, then y = y(x) or x = x(y) can be used to describe the curve. Another
method is to use coordinate transformation, e.g. using cylindrical or spherical coordinates.
In some cases, the path may need to be partitioned into several non-overlapping curves
in order to facilitate computation. One requirement is that each partition should be smooth.
1
A more general formulation would be to let the parameter start at t = a and end with t = b, where
b > a. However, using translation and scaling, it could be reduced back to a = 0 and b = 1.
146 c 2006 Tomas Co, all rights reserved
°

Example 5.1

For the closed curve shown in Figure 5.3,

d
0

−1

−2 a c
y

−3

−4
b

−5
−8 −7 −6 −5 −4 −3 −2 −1 0 1 2
x

Figure 5.3: A close path of integration in counterclockwise direction.

we have a closed elliptical curve described by


µ ¶2
x+3
+ (y + 2)2 = 4 (5.5)
2
which starts at point a : (−7, −2) then moves through point b : (−3, −4) then point c : (1, −2)
then point d : (−3, 0) and back to point a. We have the following possible descriptions:

1. Parameterized Form. First, simplify the equation to allow for easier parameteriza-
tions. For instance, we can transform (5.5) into an equation of a unit circle by using
the following variables:
x+3 y+2
X= and Y =
4 2
Then (5.5) becomes
X2 + Y 2 = 1
which we can parameterized as

X = − cos (2πt) and Y = − sin (2πt)

Note that we have included the negative signs so that the curve will start at point a
and then move counterclockwise through points b, c and d and the back to a as the
c 2006 Tomas Co, all rights reserved
° 147

value of t changes from 0 to 1. Thus, the parameterized curve, in terms of the original
variables, is given by:

Path Cabcda : x = −3 − 4 cos(2πt)


y = −2 − 2 sin(2πt)
from t = 0 to t = 1

2. Explicit function of x. We need to partition the curve into two segments. The first
segment is the lower half and the second segment is the upper half. However, since the
direction is crucial we need to indicate the start and end values of x. Thus we have for
our second description of the path:

Path Cabcda = Cabc + Ccda

where s µ ¶2
x+3
Cabc : y = −2 − 4− from x = −7 to x = 1
2
s µ ¶2
x+3
Ccda : y = −2 + 4− from x = 1 to x = −7
2

3. Explicit function of y. We need to partition the curve into three segments. The
first segment is the lower left half, the second segment is the right half and the third
segment is the upper left half. We need at least three segments to preserve the start
and end points. Thus, we have for our third description of the path:

Path Cabcda = Cab + Cbcd + Cda

where q
Cab : x = −3 − 2 4 − (y + 2)2 from y = −2 to y = −4
q
Cbcd : x = −3 + 2 4 − (y + 2)2 from y = −4 to y = 0
q
Cda : x = −3 − 2 4 − (y + 2)2 from y = 0 to y = −2

♦♦♦
148 c 2006 Tomas Co, all rights reserved
°

5.1.2 Computation of Line Integrals


Depending on computational convenience, one could choose to use a path described in either
the parameterized form or the explicit function in x or y. However, in case the path does
not lie in a flat plane, the parameterized form is usually the preferred method.
Option 1: Using Parameterized Form. In this case, the integrand is first transformed
to be a function of the parameter t, i.e. with path C : x = x(t), y = y(t), z = z(t),

f (x(t), y(t), z(t)) = g(t) (5.6)

Second, the differentials are also transformed to be in terms of dt, i.e.


dx
dx = dt (5.7)
dt
dy
dy = dt (5.8)
dt
dz
dz = dt (5.9)
dts
µ ¶2 µ ¶2 µ ¶2
dx dy dz
ds = dt + + (5.10)
dt dt dt

Finally, since t varies from 0 to 1 in a straight path, the line integral now reduces to the
ordinary integral, i.e.
1 µ ¶
dx
Z Z
f (x, y, z)dx = g(t) dt (5.11)
C 0 dt
Z 1µ ¶
dy
Z
f (x, y, z)dy = g(t) dt (5.12)
C 0 dt
Z 1µ ¶
dz
Z
f (x, y, z)dz = g(t) dt (5.13)
C 0 dt
 sµ ¶ 
Z 1 2 µ ¶2 µ ¶2
dx dy dz 
Z
f (x, y, z)ds = g(t) + + dt (5.14)
C 0 dt dt dt

Option 2: Using Explicit Function Form. As mentioned above, this approach is used
only if the path or subpath lie in flat planes. So let us just consider only the situation where
C is in the (x, y) plane. This option is just a special case of option 1. When one can describe
the path of integration with y as an explicit function of x, i.e. y = y(x), then choose a
parameterization such as
x=t and y = y(x) = y(t)
and we get back option 1. The only difference is that instead of starting at t = 0, we start
at t = xstart . Likewise, instead of ending at t = 1, we end at t = xend .
c 2006 Tomas Co, all rights reserved
° 149

Thus, we have
Z Z xend
f (x, y)dx = f (x, y(x))dx (5.15)
C
Zxstart
xend
dy
Z
f (x, y)dy = f (x, y(x)) dx (5.16)
C xstart dx
s ¶2
xend µ
dy
Z Z
f (x, y)ds = ± f (x, y(x)) 1+ dx (5.17)
C xstart dx

Likewise, if x can be described as an explicit function of y, i.e. x = x(y), we have


Z yend
dx
Z
f (x, y)dx = f (x(y), y) dy (5.18)
C ystart dy
Z Z yend
f (x, y)dy = f (x(y), y)dy (5.19)
C ystart
s µ ¶2
Z yend
dx
Z
f (x, y)ds = ± f (x(y), y) 1 + dy (5.20)
C ystart dy

The choice for positive or negative sign in the formulas given in (5.17) and (5.20)
depend on whether the projection of ds is in the same direction as dx or dy. This will be
shown in the following example.

Example 5.2

Consider the scalar function given by

f (x, y) = 2x + y + 3

and the elliptical path of integration given in Example 5.1.


Based on the parameterized form of the path of integration, as t goes from 0 to 1,

x(t) = −3 − 4 cos (2πt)


y(t) = −2 − 2 sin (2πt)

and

dx = 8π sin (2πt) dt
dy = −4π cos (2πt) dt
q
ds = 4π 4 sin2 (2πt) + cos2 (2πt)dt
150 c 2006 Tomas Co, all rights reserved
°

Thus,
Z Z 1
f (x, y)dx = (2 (−3 − 4 cos (2πt)) + (−2 − 2 sin (2πt)) + 3) (8π sin (2πt)) dt
C 0
= −8π
Z Z 1
f (x, y)dy = (2 (−3 − 4 cos (2πt)) + (−2 − 2 sin (2πt)) + 3) (−4π cos (2πt)) dt
C 0
= Z
16π
Z 1
f (x, y)ds = (2 (−3 − 4 cos (2πt)) + (−2 − 2 sin (2πt)) + 3)
C 0
µ q ¶
2 2
× 4π 4 sin (2πt) + cos (2πt) dt
= −96.885

When we use the explicit function formulation, e.g. y = y(x), for describing the path
of integration with
C = Cabc + Ccda
s µ ¶2
x+3
Cabc : y = yabc = −2 − 4 − from x = −7 to x = 1
2
s µ ¶2
x+3
Ccda : y = ycda = −2 + 4 − from x = 1 to x = −7
2
The integrand and differentials for the subpaths (denoted by the subscripts abc or cda) are
given by:
 s 
µ ¶2
x+3 
f (x, y)abc = 2x + 3 + −2 − 4 −
2
 s 
µ ¶2
x+3 
f (x, y)cda = 2x + 3 + −2 + 4 −
2
µ ¶
dy x+3
= − p
dx abc 2 (1 − x) (x + 7)
µ ¶
dy x+3
= + p
dx cda 2 (1 − x) (x + 7)
µ ¶
ds
q
2
= 1 + dyabc
dx abc
µ ¶
ds
q
2
= − 1 + dycda
dx cda
c 2006 Tomas Co, all rights reserved
° 151

Note that we used the negative sign for ds in the subpath [cda] since the direction of ds is
opposite that of dx in this region.
Using these evaluations the line integrals will be given by
Z Z 1 Z −7
f (x, y)dx = f (x, y)abc dx + f (x, y)cda dx
C −7 1
= −8π
Z 1 µ ¶ Z −7 µ ¶
dy dy
Z
f (x, y)dy = f (x, y)abc dx + f (x, y)cda dx
C −7 dx abc 1 dx cda
= Z
16π
1 µ ¶ Z −7 µ ¶
ds ds
Z
f (x, y)ds = f (x, y)abc dx + f (x, y)cda dx
C −7 dx abc 1 dx cda
= −96.885

This example shows that using either approach, the value of line integrals will be the
same. The choice is usually determined by the tradeoffs between the complexity of the
parameterization procedure and the complexity of the resulting integral.

♦♦♦

Since we will be dealing with path of integration that are either closed or not, we will
use the following notation to indicate that the line integral has a closed contour:
I
f ds C is a closed, sectionally smooth, nonintersecting path
C

The calculation procedures remain the same.

5.2 Surface Integrals


Surface integrals are specified by three components:

1. The surface of integration, S(x, y, z), which is a sectionally smooth differentiable con-
tinuous surface in (x, y, z).

2. The integrand, F (x, y, z), which is a scalar function.

3. The differential area, which could either be dxdy, dydz, dxdz or dS where dS is the
differential element of the surface of integration
152 c 2006 Tomas Co, all rights reserved
°

Based on these three components, the surface integrals are defined as the limit of a
summation of the products of the integrand F (x, y, z) and one of the differential elemental
areas, ∆S, ∆x∆y, etc. as x, y and z vary along the surface of integration.
Z Xn
F (x, y, z)dxdy = lim F (xi , yi , zi )∆xi ∆yi (5.21)
S ∆xi →0,n→∞
i=1
( surface integral with respect to (x,y) plane.
along the surface of integration S )

Z n
X
F (x, y, z)dS = lim F (xi , yi , zi )∆Si (5.22)
S ∆si →0,n→∞
i=1
( surface integral with respect to the surface domain S
along the surface of integration S )
R R
Similar definitions apply to S F (x, y, z)dydz and S F (x, y, z)dxdz.
To help us visualize an interpretation of surface integrals, we could go back to the
scenario of the creature that was mining for a substance A in the three dimensional ground.
This time, the scalar function
R f (x, y, z) could be imagined as the amount of A per unit area.
Then the surface integral, S f (x, y, z)dS, becomes the total amount of A mined by this
creature as it sweeps a surface area S out of the ground. Note that the surface S will not
necessarily be a flat plane.

5.2.1 Surface of Integration


A general description of smooth and differentiable three dimensional surface is given by the
form
φ(x, y, z) = 0
where φ is a general differentiable nonlinear function. However, since the surface is necessarily
two dimensional regardless of whether the surface is flat or curved, it is often described by
the use of two independent parameters, say u and v. This means that the surface can usually
be described as follows:
S : (x(u, v), y(u, v), z(u, v)) as u and v vary independently in a closed domain.
(5.23)
In some cases, the surface can be described by letting u = x and v = y. If this is so,
then the surface can be described as an explicit function for z
S : z = z(x, y) as x and y vary independently in a close domain (5.24)

From the parameterized description of the surface, i.e. S : (x(u, v), y(u, v), z(u, v)), we
can evaluate two important variables. One variable is the direction of a vector normal to the
c 2006 Tomas Co, all rights reserved
° 153

surface at a point (x(uo , vo ), y(uo , vo ), z(uo , vo )). The other variable is the differential area,
dS, which is related to the product dudv (note that du and dv are not necessarily orthogonal
to each other). Fortunately, these two variables can be determined at the same time.
Recall that the position vector

R = xδ x + yδ y + zδ z (5.25)

is the vector starting at the origin and ending at the point (x, y, z). As the parameter u
is varied while v is fixed, a path, say C1 , is traced. Likewise, as the parameter v is varied
while parameter u is held fixed, (x(uo , v), y(uo , v), z(uo , v)) will trace out another path, say
C2 . This is shown in Figure 5.4. From the same figure, we see that the differential change
in R tangential to the path C1 is given by
µ ¶
∂R ∂x ∂y ∂z
a = du = du δ + δ + δ (5.26)
∂u ∂u x ∂u y ∂u z

while the differential change in R tangential to the path C2 is given by


µ ¶
∂R ∂x ∂y ∂z
b = dv = dv δ + δ + δ (5.27)
∂v ∂v x ∂v y ∂v z

Figure 5.4: Graphical representation of differential area, dS, and the normal vector, N.

Then a vector perpendicular to both a and b will be the cross product of a and b, i.e.

N = a×b
154 c 2006 Tomas Co, all rights reserved
°

¯ ¯
¯
¯ δ x δ y δ z
¯
¯
¯ ¯
¯ ¯
= ¯¯ ∂x/∂u ∂y/∂u ∂z/∂u ¯ dudv
¯
¯ ¯
¯ ¯
¯ ∂x/∂v ∂y/∂v ∂z/∂v ¯
µ ¶
∂(y, z) ∂(z, x) ∂(x, y)
= δ + δ + δ dudv (5.28)
∂(u, v) x ∂(u, v) y ∂(u, v) z
where, ¯ ¯
¯ ∂a/∂c ∂a/∂d ¯
∂(a, b) ¯¯ ¯
=¯ ¯
∂(c, d) ¯ ¯
∂b/∂c ∂b/∂d ¯
is a shorthand notation for the Jacobian determinant.
To obtain the unit normal vector, we simply divide by the norm, i.e.
N
n=
kNk
This also removes the dependency on the factor dudv.
From the previous chapters, recall that the magnitude of cross product of the two
vectors evaluates the area of the parallelogram formed by these two vectors. Thus the
differential area dS is given by
dS = kNk
sµ ¶2 µ ¶2 µ ¶2
∂(y, z) ∂(z, x) ∂(x, y)
= + + dudv (5.29)
∂(u, v) ∂(u, v) ∂(u, v)

If the surface of integration can be described by z = z(x, y), we let u = x and v = y.


This reduces the formulas above to
µ ¶
∂z ∂z
N = δ − δ + δ z dxdy (5.30)
∂x x ∂y y
³ ´
∂z ∂z
δ
∂x x
− δ
∂y y
+ δ z
n = r ³ ´2 (5.31)
¡ ∂z ¢2 ∂z
1 + ∂x + ∂y
s 
µ ¶2 µ ¶2
∂z ∂z 
dS =  1 + + dxdy (5.32)
∂x ∂y

The choice for sign depends on the interpretation of the surface direction. For a surface
that encloses a region of three dimensional space, the surface outward of the enclosed region
is often given a positive sign.
c 2006 Tomas Co, all rights reserved
° 155

5.2.2 Computation of Surface Integrals


With the parameterization of the surface using the parameters u and v, the domain is a
closed two dimensional plane in the (u, v) space.
In one case, the domain is described by independent ranges of u and v. For instance,
u and v could have been scaled to be between 0 and 1, i.e. domain D is given as

D : 0≤u≤1 ; 0≤v≤1 (5.33)

If the domain can be described as (5.33), then the surface integral can be computed
as follows: Z Z 1Z 1
f (x, y, z)dS = g(u, v)dudv (5.34)
S 0 0

where
sµ ¶2 µ ¶2 µ ¶2
∂(y, z) ∂(z, x) ∂(x, y)
g(u, v) = f (x(u, v), y(u, v), z(u, v)) + + (5.35)
∂(u, v) ∂(u, v) ∂(u, v)

and the computation proceeds to two sequential steps:

1. Step 1. Calculate h(v), where


Z 1
h(v) = g(u, v)du holding v constant while integrating
0

2. Step 2. Z Z 1
f (x, y, z)dS = h(v)dv
S 0

In more general cases, the boundaries of u and v may be interdependent. If this is the
case, two descriptions of the domain may be used:

Du : ulower ≤ u ≤ uupper ; φ0 (u) ≤ v ≤ φ1 (u) (5.36)


or
Dv : vlower ≤ v ≤ vupper ; ψ0 (v) ≤ u ≤ ψ1 (v) (5.37)

where ulower , uupper , vlower and vupper are constants. Both domain descriptions are shown in
Figure 5.5 are equally valid. The choice usually depends on which description will yield more
efficient computation.
Using the first description given in (5.36), the computation proceeds in two sequential
steps, using g(u, v) in (5.35),
156 c 2006 Tomas Co, all rights reserved
°

Figure 5.5: The two possible domain descriptions: (a) boundary is partitioned into two
segments such that v = φ(u), and (b) boundary is partitioned into two segments such that
u = ψ(v) .

1. Step 1. Calculate h(u), where


Z φ1 (u)
h(u) = g(u, v)dv holding u constant while integrating
φ0 (u)

2. Step 2. Z Z uupper
f (x, y, z)dS = h(u)du
S ulower

Using the second description given in (5.37), the computation also proceeds in two
sequential steps:

1. Step 1. Calculate h(v), where


Z ψ1 (v)
h(v) = g(u, v)du holding u constant while integrating
ψ0 (v)

2. Step 2. Z Z vupper
f (x, y, z)dS = h(v)dv
S vlower
c 2006 Tomas Co, all rights reserved
° 157

In the special case that the surface is given explicitly by z = z(x, y), we use the
following substitutions to the steps provided above,

u = x
v = y
s µ ¶2 µ ¶2
∂z ∂z
g(u, v) = g(x, y) = f (x, y, z(x, y)) 1+ +
∂x ∂y
ulower = xlower uupper = xupper
φ1 (u) = φ1 (x) φ0 (u) = φ0 (x)
vlower = ylower vupper = yupper
ψ1 (v) = ψ1 (y) ψ0 (v) = ψ0 (y)

Example 5.3

We wish to find the surface integral of the function given by

f (x, y, z) = 2x + y − z + 3

on the surface given by the ellipse:


³ y ´2
x2 + + z2 = 1
2

We could parameterize the surface as follows:

z = cos(u)
x = sin(u) cos(v)
y = sin(u) sin(v)

with the given domain


0 ≤ u ≤ 2π 0≤v≤π
The Jacobian determinants can be found to be:
∂(x, y)
= −2 sin(u) cos(u)
∂(u, v)
∂(y, z)
= −2 sin2 (u) cos(v)
∂(u, v)
∂(x, z)
= sin2 (u) cos(v)
∂(u, v)
158 c 2006 Tomas Co, all rights reserved
°

and this gives us


sµ ¶2 µ ¶2 µ ¶2
∂(y, z) ∂(z, x) ∂(x, y)
g(u, v) = f (x, y, z) + +
∂(u, v) ∂(u, v) ∂(u, v)
= α(u, v)β(u, v)

where

α(u, v) = 2 sin(u) cos(v) + 2 sin(u) sin(v) − cos(u) + 3


q
3 cos2 (v) (cos(u) − 1)2 (cos(u) + 1)2 + (1 + 2 cos2 (u) − 3 cos4 (u))
¡ ¢
β(u.v) =

The surface integral can then be solved numerically to be


Z π Z 2π
g(u, v)dudv = 64.4
0 0

As an alternative, we partition the elliptical surface into two halves. The upper half
can be described by the function
s µ ¶
2
y2
zu = 1 − x −
2

while the lower half is described by


s µ ¶
y2
zℓ = − 1− x2 −
2

In either half, the domain can be described by


√ √
D : −1≤x≤1 − 2 1 − x2 ≤ y ≤ 2 1 − x2

For the upper half,


dzu −2x
= p
dx 4 − 4x2 − y 2
dzu −y/2
= p
dy 4 − 4x2 − y 2
which then makes the integrand
às !às !
3y 2 − 16
µ ¶
y2 1
gu (x, y) = 2x + y − 1 − x2 − +3
2 2 −4 + 4x2 + y 2
c 2006 Tomas Co, all rights reserved
° 159

For the lower half,

dzℓ 2x
= p
dx 4 − 4x2 − y 2
dzℓ y/2
= p
dy 4 − 4x2 − y 2

which then makes the integrand


às !às !
3y 2 − 16
µ ¶
y2 1
gℓ (x, y) = 2x + y + 1 − x2 − +3
2 2 −4 + 4x2 + y 2

Combining everything, we can calculate the surface integral via numerical integration
to be


Z +1 Z 2 1−x2
Iu = √
gu (x, y)dydx = 26.6
−1 −2 1−x2

Z +1 Z 2 1−x2
Iℓ = √
gℓ (x, y)dydx = 37.8
−1 −2 1−x2
Z
f dS = Iu + Iℓ = 64.4
S

which is the same value as the previous answer using the parameterized description.

♦♦♦

5.3 Volume Integrals


Volume integrals are one-dimensional extensions to surface integrals. They involve three
components:

1. The volume of integration, V (x, y, z), which is a closed three dimensional region
bounded by sectionally smooth differentiable surfaces.

2. The integrand, F (x, y, z), which is a scalar function.

3. The differential area, given by dxdydz


160 c 2006 Tomas Co, all rights reserved
°

Based on these three components, the volume integrals are defined as the limit of a
summation of the products of the integrand F (x, y, z) and the differential elemental volume,
∆V = ∆x∆y∆z, which varies along the volume of integration.
Z Xn
F (x, y, z)dxdydz = lim F (xi , yi , zi )∆xi ∆yi ∆zi (5.38)
V ∆Vi →0,n→∞
i=1

To continue the visual interpretation tool used for line and surface integrals, we go back
to the scenario of the creature that was mining for a substance A in the three dimensional
ground. This time, the scalar function F (x,
R y, z) could be imagined as the amount of A per
unit volume. Then the volume integral, V F (x, y, z)dV , becomes the total amount of A
mined by this creature as it covers the region V carved out of the ground.

5.3.1 Volume of Integration


In some cases, another set of coordinates might allow for easier computation, e.g. cylindrical
or spherical coordinates. In general, this set of new coordinates could be seen as parameters
defining x, y and z,

x = x(u, v, w) (5.39)
y = y(u, v, w) (5.40)
z = z(u, v, w) (5.41)

where u, v and w are the new coordinates or parameters.


To convert the differential volume dV in terms of dudvdw, we can use the same proce-
dures used when obtaining dS for surface integrals. First, recall the definition of the position
vector, R,
R = xδ x + yδ y + zδ z
By tracing changes in R as we vary u while keeping v and w constant, a curve can be
generated inside the region V . Let us denote this curve as C1 . Likewise, we could trace two
more curves. Denote by C2 , the curve that is generated by tracing R as we vary v while
keeping u and w constant. Also, denote by C3 , the curve that is generated by tracing R
as we vary w while keeping u and v constant. As we can see from Figure 5.6, a differential
change along each of these curves will yield three non-coplanar vectors:
µ ¶ µ ¶
∂R ∂x ∂y ∂z
a= du = δ + δ + δ du (5.42)
∂u ∂u x ∂u y ∂u z
µ ¶ µ ¶
∂R ∂x ∂y ∂z
b= dv = δ + δ + δ dv (5.43)
∂v ∂v x ∂v y ∂v z
µ ¶ µ ¶
∂R ∂x ∂y ∂z
c= dw = δ + δ + δ dw (5.44)
∂w ∂w x ∂w y ∂w z
c 2006 Tomas Co, all rights reserved
° 161

Figure 5.6: Graphical representation of differential volume, dV , as function of u, v and w.

The differential volume is the volume occupied by the parallelepiped formed by the
vectors a, b and c. The volume can be computed by taking the absolute value of the triple
product of these three vectors, i.e.
dV = |c · (a × b)| = |a · (b × c)|
¯µ ¶ µ ¶¯
¯ ∂R ∂R ∂R ¯¯
= ¯
¯ · × dudvdw
∂u ∂v ∂w ¯
¯µ ¶¯
¯ ∂(x, y, z) ¯
= ¯¯ ¯ dudvdw (5.45)
∂(u, v, w) ¯
where we used the shorthand notation for the Jacobian determinant
 
∂x/∂u ∂x/∂v ∂x/∂w
 
∂(x, y, z)  
= det  ∂y/∂u ∂y/∂v ∂y/∂w  (5.46)
∂(u, v, w) 



∂z/∂u ∂z/∂v ∂z/∂w

5.3.2 Computation of Volume Integrals


Having determined the differential volume and the integrand, one needs to identify the limits
of integration in each of the variables x, y and z, or equivalently, of parameters u, v and w.
In the simpler case in which the limits are independent, i.e. if the limits of the bound-
aries are given by
umin ≤ u ≤ umax (5.47)
162 c 2006 Tomas Co, all rights reserved
°

vmin ≤ v ≤ vmax (5.48)


wmin ≤ w ≤ wmax (5.49)

the volume integral can be integrated in a nested fashion by the following


Z wmax Z vmax Z umax ¯ ¯
¯ ∂(x, y, z) ¯
Z
F dV = F (x(u, v, w), y(u, v, w), z(u, v, w)) ¯¯ ¯ dudvdw
V wmin vmin umin ∂(u, v, w) ¯
(5.50)
For the more general case where the variables are interdependent when describing the
surface boundaries, there are six possible descriptions that can be used based on the sequence
of dependencies. We will describe the sequence w → v → u, but the other descriptions are
very similar.
As shown in Figure 5.7, we can first identify the maximum and minimum value that
z can take, i.e.
wmin ≤ w ≤ wmax (5.51)

Figure 5.7: A nested description of volume boundaries.

Next, we take a slice of the volume by fixing w that belongs to the allowed range.
This slice will be an enclosed curve for which we can identify also a maximum and minimum
value for v,
ηmin (w) ≤ v ≤ ηmax (w) (5.52)
Finally, we can see from Figure 5.7 that the limits of v for this slice will divide the closed
curve into two segments. Each of these segments can then be described by functions of v
and w (that were fixed for this slice), i.e.

ξmin (v, w) ≤ u ≤ ξmax (v, w) (5.53)


c 2006 Tomas Co, all rights reserved
° 163

Thus we end up with a slightly different nested integration given by


Z wmax Z ηmax (w) Z ξmax (v,w) ¯ ¯
¯ ∂(x, y, z) ¯
Z
F dV = F¯
¯ ¯ dudvdw (5.54)
V wmin ηmin (w) ξmin (v,w) ∂(u, v, w) ¯

Example 5.4

Consider the volume of integration given by the ellipse


³ y ´2
2
x + + z2 ≤ 1
2
and the integrand given by
F (x, y, z) = 2x + y − z + 3

Using the following parameterization:

x = u sin(v) cos(w)
y = 2u sin(v) sin(w)
z = u cos(v)

where the boundaries of the variables are given as

0≤ u ≤1
0 ≤ v ≤ 2π
0≤ w ≤π

The differential volume is then given by


¯  ¯
¯
¯ sin(v) cos(w) u cos(v) cos(w) −u sin(v) sin(w) ¯
¯
¯  ¯
¯  ¯
dV = ¯¯det   2 sin(v) sin(w) 2u cos(v) sin(w) 2u sin(v) cos(w)
¯ dudvdw
¯
¯  ¯
¯ ¯
¯ cos(v) −u sin(v) 0 ¯
= 2u2 |sin(v)| dudvdw

while the integrand becomes

F = 2u sin(v) (cos(w) + sin(w)) − u cos(v) + 3

Combining all the elements together, we can compute the volume integral as
Z π Z 2π Z 1
(2u sin(v) (cos(w) + sin(w)) − u cos(v) + 3) 2u2 |sin(v)| dudvdw = 8π
¡ ¢
0 0 0
164 c 2006 Tomas Co, all rights reserved
°

As an alternative, we could use the original variables x, y and z. Doing so, we have
the differential volume as dV = dxdydz.
The boundaries of the volume of integration can be described by
r ³ y ´2 r ³ y ´2
Surface boundary : − 1− z2 − ≤ x ≤ 1 − z2 −
2 2
√ √
−2 1 − z 2 y 2 1 − z2
−1 ≤ z ≤1

Thus the volume integral is given by


Z 1 Z √
2 1−z 2 Z √1−z2 −(y/2)2
√ √ (2x + y − z) dxdydz = 8π
−1 −2 1−z 2 − 1−z 2 −(y/2)2

which is the same answer obtained by using the parameterized description.

♦♦♦

5.4 Green’s Lemma


We begin with one of the basic formulas that consists of the relationship between a surface
integral on surface S and a line integral on closed curve C that is the boundary of S. We
will state this formula in terms of parameters u and v to allow for the application of the
formulas to curved surfaces in three dimensional space.

Lemma 5.1 Green’s Lemma: Let F (u, v) and G(u, v) be differentiable functions in the
domain D ∈ R2 . Then
Z µ ¶
∂G ∂F
I
(F (u, v)du + G(u, v)dv) = − dudv (5.55)
C S ∂u ∂v

where

1. The closed curve C is the boundary of the surface of integration S, which should be
sectionally smooth.

2. The positive direction of contour C is consistent with the definition of the positive
direction of the vector: N = [(∂R/∂u) × (∂R/∂v)], where R is the position vector.
c 2006 Tomas Co, all rights reserved
° 165

For the special case in which the surface of integration S lies in the (x, y)-plane,
Green’s lemma is stated in terms of x and y, i.e. (5.55) is replace by a formula with u = x
and v = y. In this case, the positive direction of the contour is conventionally set to be
counterclockwise. Equivalently, based on the usual arrangements of the x axis and the y
axis, the positive contour direction is chosen such that the region in S is always to the left
of the contour’s path.

Proof of Green’s lemma. To prove the lemma, we make use of two possible
descriptions of the boundary as given in (5.36) and (5.37).
Recalling (5.36), the domain of the surface of integration S is given by
D : ulower ≤ u ≤ uupper ; φ0 (u) ≤ v ≤ φ1 (u)
where the closed contour C is the sum given by
C = C0,v − C1,v
where C0,v and C1,v are the curves described by φ0 (u) and φ1 (u), respectively,
are positive with increasing values of u.
Applying this description to the second surface integral in (5.55),
Z uupper ÃZ φ1 (u) !
∂F (u, v) ∂F (u, v)
Z
dudv = dv du
S ∂v ulower φ0 (u) ∂v
Z uupper
= (F (u, φ1 (u)) − F (u, φ0 (u))) du
ulower
I
= − F (u, v)du (5.56)
C

Likewise, using (5.37), the domain of the surface of integration S is given by


D : vlower ≤ v ≤ vupper ; ψ0 (v) ≤ u ≤ ψ1 (v)
where the closed contour C is now equal to the sum given by
C = C1,u − C0,u
where C0,v and C1,v are the curves described by ψ0 (u) and ψ1 (u), respectively,
are positive with increasing values of v.
Applying this domain description to the first surface integral in (5.55),
Z vupper ÃZ ψ1 (v) !
∂G(u, v) ∂G(u, v)
Z
dudv = dv du
S ∂u vlower ψ0 (v) ∂u
Z vupper
= (G(ψ1 (v), v) − G(ψ0 (v), v)) du
vlower
I
= G(u, v)dv (5.57)
C
166 c 2006 Tomas Co, all rights reserved
°

Combining (5.56) and (5.57), we arrive at the formula given in Green’s lemma,
∂G(u, v) ∂F (u, v)
I I Z Z
G(u, v)dv + F (u, v)du = dudv − dudv
C C S ∂u S ∂v
QED

Example 5.5

We want to verify Green’s lemma for the functions F (x, y, z) and G(x, y, z) given by
F (x, y, z) = x2 + y
G(x, y, z) = 2x + 3y − z + 5
Suppose further that the surface of integration to be the top portion of the unit sphere with
0 ≤ θ ≤ π/4 and 0 ≤ φ ≤ 2π as shown in Figure 5.8.

Figure 5.8: The surface of integration given by the top of a unit sphere.

One parameterization of the variables, with u = φ and v = θ, is given by


x = sin(v) cos(u)
y = sin(v) sin(u)
z = cos(v)

Substituting, we get
F (u, v) = (sin(v) cos(u))2 + sin(v) sin(u)
G(u, v) = 2 sin(v) cos(u) + 3 sin(v) sin(u) − cos(v) + 5
whose partial derivatives are given by
∂F
= cos(v) sin(u) + 2 sin(v) cos2 (u)
¡ ¢
∂v
∂G
= sin(v) (3 cos(u) − 2 sin(u))
∂u
c 2006 Tomas Co, all rights reserved
° 167

The surface integral in the lemma can now be evaluated to be


π/4 2π µ ¶
∂G ∂F π
Z Z
− dudv = −
0 0 ∂u ∂v 2

Since the chosen parameterization was for u = φ and u = θ, the closed contour in the
(u, v) plane is shown in Figure 5.9. The positive direction is counterclockwise to yield an
vector that is normal to S that is outward from the center of the sphere.

Figure 5.9: The closed path of integration in the (u, v) plane.

Based on Figure 5.9, the line integrals can be calculated to be


I Z 2π Z 0
³ π´
F (u, v)du = F (u, 0) du + F u, du
C 0 2π 4
π
= 0 −
2

I Z π/4 Z 0
G(u, v)dv = G (2π, v) dv + G (0, v) dv
C 0 π/4
à √ ! à √ !
3 2 π 3 2 π
= − +2+5 + −2−5
2 4 2 4

Combining all the results, we arrive at


Z µ ¶
∂G ∂F
I I
F du + Gdv = − dudv
C C S ∂u ∂v

♦♦♦
168 c 2006 Tomas Co, all rights reserved
°

5.4.1 Surfaces with Holes


R
From example 5.5, the computation of C G(u, v)dv resulted in zero. Two of the path seg-
ments involved the same path except that they were directed opposite of each other. As
expected, the line integrals corresponding to both these paths canceled each other out.
We will use this observation to allow us to apply Green’s lemma to surfaces that have
holes in them. For illustration purposes, consider the surface shown in Figure 5.10b.
This surface, S1 , was obtained from the original surface S after another surface S2 was
removed. In Figure 5.10c, a continuous contour C1 can be generated to be the boundary of
S1 . At some point along the outer boundary, contour C1 cuts a path to reach the inner hole.
Then the contour will traverse the boundary of this hole. After reaching the point where the
contour first entered the hole, the contour C1 will retrace the same cutting path backwards
to continue tracking the outer path.
With this technique, the two line integrals that have traversed the same path segments,
but in opposite directions, will cancel each other out. Thus, as shown in Figure 5.10d, the
line integral along C1 will be equal to the line integral along C minus the line integral along
C2 , which is the contour of the removed surface S2 , i.e.
I I I
ωds = ωds − ωds (5.58)
C1 C C2

From the definition of surface integrals, we know that


Z Z Z
ΩdS = ΩdS − ΩdS (5.59)
S1 S S2

Applying Green’s lemma and (5.58) to (5.59),


Z µ ¶
∂G ∂F
I I
− dudv = (F du + Gdv) − (F du + Gdv)
S1 ∂u ∂v C C2
I
= (F du + Gdv) (5.60)
C1

Equation (5.60) implies that the strategy of cutting through the region to connect the outer
contour with the inner contours of the holes validates the applicability of the formula of
Green lemma to surfaces with holes.

5.5 Divergence Theorem


Consider a vector field F. The divergence theorem states that the volume integral of the
divergence ∇ · F in a given volume region V can also be evaluated indirectly by taking the
surface integral of the flux of F, i.e. F · n, over the surface S that bounds the region V .
c 2006 Tomas Co, all rights reserved
° 169

Figure 5.10: The surface of integration with holes: (a) original surface S and contour C,
(b) S1 obtained by removing S2 , (c) obtaining continuous contour
H C
H 1 forH S1 , (d) due to the
cancelation of line integrals in oppositely directed segments, C1 = C − C2 .
170 c 2006 Tomas Co, all rights reserved
°

Theorem 5.1 Let F(x, y, z) be a vector field which is differentiable in V then


Z Z
F · n dS = ∇ · F dV (5.61)
S V

Proof of Divergence Theorem:

In rectangular coordinates, let F be given by

F = Fx δ x + Fy δ y + Fz δ z

The volume integral in (5.61) can be expanded to be the sum of three terms
∂Fx ∂Fy ∂Fz
Z Z Z Z
∇ · F dV = dV + dV + dV (5.62)
V V ∂x V ∂y V ∂z

For the first term in (5.62), we can use the following description of the volume
of integration: 2

V : zmin ≤ z ≤ zmax
ηmin (z) ≤ y ≤ ηmax (z)
ξmin (y, z) ≤ x ≤ ξmax (y, z)

to obtain the following triple integral formulation


Z zmax Z ηmax (z) Z ξmax (y,z)
∂Fx ∂Fx
Z
dV = dxdydz
V ∂x zmax ηmin (z) ξmin (y,z) ∂x

After performing the inner integration with respect to x, the result is a difference
of two surface integrals
Z zmax Z ηmax (z)
∂Fx
Z
dV = Fx (ηmax (y, z), y, z) dydz
V ∂x zmin ηmin (z)
Z zmax Z ηmax (z)
− Fx (ηmin (y, z), y, z) dydz (5.63)
zmax ηmin (z)

The first surface integral in (5.63) is with respect to the surface: S1 : x =


ξmax (y, z). To determine the differential area of the surface, dS1 at a point in
2
This assumes that any line that is parallel to the x axis will intersect the surface boundary of region V
at two points, except at the edges of the boundary where it touches at one point. If this assumption is not
true for V , it can always be divided into subsections for which this assumption can hold. After applying the
divergence theorem to these smaller regions, they can be added up later and the resulting sum can be shown
to satisfy the divergence theorem.
c 2006 Tomas Co, all rights reserved
° 171

the surface, we can use the position vector R of the point in surface S1 . Along
the curve in the surface, in which z is fixed, we have a tangent vector given by
(∂R/∂y) dy. Likewise, along the curve in the surface, in which y is fixed, we have
a tangent vector given by (∂R/∂z)dz. This is shown in Figure 5.11. By taking
the cross product of these two tangent vectors, we obtain a vector N1 which is
normal to surface S1 whose magnitude is the area of the parallelogram formed
by the two tangent vectors, i.e.

N1 = dS1 n1

where n1 is the unit normal vector.

Figure 5.11: The normal vector to the surface x = ξmax (y, z) is given by N1 which has a
magnitude equal to the differential surface, dS1 .

Thus, with the position vector R along the surface given by

R = ξmax (y, z) δ x + y δ y + z δ z

we have
µ ¶
∂R ∂R
dS1 n1 = × dydz
∂y ∂z
µ ¶ µ ¶
∂ξmax ∂ξmax
= δx + δy × δ x + δ z dydz
∂y ∂z
µ ¶
∂ξmax ∂ξmax
= δx − δy − δ z dydz
∂z ∂y
By taking the dot product of both sides with δ x ,

(n1 · δ x ) dS1 = dydz (5.64)


172 c 2006 Tomas Co, all rights reserved
°

The same arguments can be used for the other surface given by x = ξmin (y, z).
The difference is that, as shown in Figure 5.12, the normal vector N2 =
(∂R/∂z) × (∂R/∂y), and thus
(n2 · δ x ) dS2 = − dydz (5.65)

Figure 5.12: The normal vector to the surface x = ξmin (y, z) is given by N2 which has a
magnitude equal to the differential surface, dS2 .

Returning to equation (5.63), we can now use the results in (5.64) and (5.65) to
obtain,

∂Fx
Z Z Z Z
dV = Fx (n1 · δ x ) + Fx (n2 · δ x ) = Fx δ x · n (5.66)
V ∂x S1 S2 S

Following the same procedure, we could show that the other two terms in (5.62)
can be evaluated to be
∂Fy
Z Z
dV = Fy δ y · n (5.67)
V ∂y S
∂Fz
Z Z
dV = Fz δ z · n (5.68)
V ∂z S

Adding the three equations: (5.66), (5.67) and (5.68), we end up with the diver-
gence theorem, i.e.

Z µ ¶
∂Fx ∂Fy ∂Fy
Z
¡ ¢
+ + dV = Fx δ x + Fy δ y + Fz δ z · n dS (5.69)
V ∂x ∂y ∂z S

QED
c 2006 Tomas Co, all rights reserved
° 173

5.6 Stokes’ Theorem


Theorem 5.2 Let F = Fx (x, y, z) δ x + Fy (x, y, z) δ y + Fz (x, y, z) δ z be a
differentiable vector field in a surface S, where S is a connected and sectionally
smooth surface in (x, y, z)-space bounded by the closed curve C. Then
Z Z
F · dR = (∇ × F) · n dS (5.70)
C S

Proof of Stokes’ Theorem:

Let the surface S be parameterized by u and v, i.e. with x = x(u, v), y = y(u, v)
and z = z(u, v). Then the differentials dx, dy and dz can be described by
∂x ∂x
dx = du + dv
∂u ∂v
∂y ∂y
dy = du + dv
∂u ∂v
∂z ∂z
dz = du + dv
∂u ∂v
The line integral in (5.70) can then be put in terms of u and v:
I I
F · dR = Fx dx + Fy dy + Fz dz
C C
µ ¶
∂x ∂x
I
= Fx du + dv
C ∂u ∂v
µ ¶
∂y ∂y
I
+ Fy du + dv
C ∂u ∂v
µ ¶
∂z ∂z
I
+ Fz du + dv
C ∂u ∂v
I
= f (u, v)du + g(u, v)dv (5.71)
C

where,
∂x ∂y ∂z
f (u, v) = Fx + Fy + Fz (5.72)
∂u ∂u ∂u
∂x ∂y ∂z
g(u, v) = Fx + Fy + Fz (5.73)
∂v ∂v ∂v
By applying Green’s lemma, i.e. (5.55), to (5.71)
Z µ ¶
∂g ∂f
I
(f (u, v)du + g(u, v)dv) = − dudv (5.74)
C S ∂u ∂v
174 c 2006 Tomas Co, all rights reserved
°

Based on f (u, v) and g(u, v) given in (5.72) and (5.73), respectively, the integrand
of the surface integral in (5.74) can be put in terms of the curl of F as follows:
c 2006 Tomas Co, all rights reserved
° 175

∂ 2x ∂ 2y
µ
∂g ∂f ∂Fx ∂x ∂Fy ∂y
− = + Fx + + Fy
∂u ∂v ∂u ∂v ∂v∂u ∂u ∂v ∂v∂u
2

∂Fz ∂z ∂ z
+ + Fz
∂u ∂v ∂v∂u
2
∂ 2y
µ
∂Fx ∂x ∂ x ∂Fy ∂y
− + Fx + + Fy
∂v ∂u ∂u∂v ∂v ∂u ∂u∂v
2

∂Fz ∂z ∂ z
+ + Fz
∂v ∂u ∂u∂v
∂Fx ∂x ∂x ∂Fx ∂y ∂x ∂Fx ∂z ∂x
= + +
∂x ∂u ∂v ∂y ∂u ∂v ∂z ∂u ∂v
∂Fy ∂x ∂y ∂Fy ∂y ∂y ∂Fy ∂z ∂y
+ + +
∂x ∂u ∂v ∂y ∂u ∂v ∂z ∂u ∂v
∂Fz ∂x ∂z ∂Fz ∂y ∂z ∂Fz ∂z ∂z
+ + +
∂x ∂u ∂v ∂y ∂u ∂v ∂z ∂u ∂v
∂Fx ∂x ∂x ∂Fx ∂y ∂x ∂Fx ∂z ∂x
− − −
∂x ∂v ∂u ∂y ∂v ∂u ∂z ∂v ∂u
∂Fy ∂x ∂y ∂Fy ∂y ∂y ∂Fy ∂z ∂y
− − −
∂x ∂v ∂u ∂y ∂v ∂u ∂z ∂v ∂u
∂Fz ∂x ∂y ∂Fz ∂y ∂y ∂Fz ∂z ∂y
− − −
∂x ∂v ∂u ∂y ∂v ∂u ∂z ∂v ∂u
∂Fx ∂(y, x) ∂Fx ∂(z, x)
= +
∂y ∂(u, v) ∂z ∂(u, v)
∂Fy ∂(x, y) ∂Fy ∂(z, y)
+ +
∂x ∂(u, v) ∂z ∂(u, v)
∂Fz ∂(x, z) ∂Fz ∂(y, z)
+ +
∂x ∂(u, v) ∂y ∂(u, v)
µ ¶ µ ¶
∂Fy ∂Fx ∂(x, y) ∂Fz ∂Fy ∂(y, z)
= − + −
∂x ∂y ∂(u, v) ∂y ∂z ∂(u, v)
µ ¶
∂Fx ∂Fz ∂(z, x)
+ −
∂z ∂x ∂(u, v)
µ
∂(y, z) ∂(z, x)
= (∇ × F) · δx + δ
∂(u, v) ∂(u, v) y

∂(x, y)
+ δ (5.75)
∂(u, v) z

Recall (5.28) for calculating N which is the vector normal to the surface S and
176 c 2006 Tomas Co, all rights reserved
°

having a magnitude equal to dS,


µ ¶
∂(y, z) ∂(z, x) ∂(x, y)
N = n dS = δ + δ + δ dudv (5.76)
∂(u, v) x ∂(u, v) y ∂(u, v) z

Combining (5.71), (5.74), (5.75) and (5.76), will result in


I Z
F · dR = (∇ × F) · ndS
C S

which is Stokes’ theorem.


QED

5.7 Path Independence of Line Integrals


Definition 5.1 Let F = Fx δ x + Fy δ y + Fz δ z be a given vector field and R =
x δ x + y δ y + z δ z be the position vector of a point (x, y, z). Then the line integral
Z
IC = F · dR (5.77)
C

is independent of path in a region V , if for any pair of curves, C1 and C2 inside


the region V ,
IC1 = IC2 (5.78)
where C1 and C2 are continuous and sectionally smooth curves that have the same
initial point (xi , yi , zi ) and the same end point (xf , yf , zf ).

If we leave out the specification of region V , then it is usually understood that the
phrase “independent of path” refers to the whole three dimensional space as the region of
interest.
Let us first consider two paths C1,AB and C2,AB in region V that does not intersect
each other except for the start and end points. If the line integrals are independent of path
in V , Z Z
F · dR = F · dR (5.79)
C1,AB C2,AB

which could be combined into one integral,


Z Z
F · dR − F · dR = 0
C1,AB C2,AB
Z
F · dR =
C1,AB −C2,AB
I
F · dR = 0 (5.80)
C
c 2006 Tomas Co, all rights reserved
° 177

where C = C1,AB − C2,AB is a simple closed path as shown in Figure 5.13. With any choice
of C1,AB and C2,AB (so far assumed to be nonintersecting at midpath), by reversing the path
direction of C2,AB , a closed path C is generated. Thus path independence guarantees that
the line integral using the closed path will have to be zero.

Figure 5.13: Two paths with the same start point A and end point B.

Alternatively, we could have rearranged (5.79) to be


Z I
F · dR = 0 = F · dR (5.81)
C2,AB −C1. AB C′

where C ′ = C2,AB − C1,AB is also a simple closed path but in the opposite direction of C.
Now consider the situation where path C1,AB and C2,AB might intersect each other
somewhere between the end points, say at point D. Path C1,AB can then be partitioned to
be the sum of two subpaths: C1,AD and C1,DB . Likewise, path C2,AB can also be partitioned
to be the sum of two subpaths: C2,AD and C2,DB . This is shown in Figure 5.14.
Equation (5.79) can now be expanded to be
Z Z
F · dR − F · dR = 0
C1,AB C2,AB
Z Z Z Z
F · dR + F · dR − F · dR − F · dR = 0
C1,AD C1,DB C2,AD C2,DB
178 c 2006 Tomas Co, all rights reserved
°

Figure 5.14: Two paths with the same start point A and end point B plus an intersection
at point D.
Z Z Z Z
F · dR − F · dR + F · dR − F · dR = 0
C1,AD C2,AD C1,DB C2,DB
Z Z
F · dR + F · dR = 0
C1,AD −C2,AD C1,DB −C2,DB
I I
F · dR − F · dR = 0 (5.82)
CAD CDB

where CAD is the closed path formed by adding path C1,AD and reverse of path C2,AD .
Similarly, CDB is the closed path formed by adding path C1,DB and reverse of path C2,DB .
Since the definition of path independence applies to the subpaths, the two closed paths will
each generate a zero line integral also.
Thus, the conditions for path independence is equivalent to having the line integral of
any closed paths in region V all be zero, including whichever direction the closed path takes.
Having established this equivalence, we can use Stokes’ theorem to offer one more tool
to help determine path independence. Recall that Stokes’ theorem relates the line integral
in a closed path to a surface integral involving the curl of vector field F:
I Z
F · dR = n · (∇ × F) (5.83)
C S

If ∇ × F = 0 for all point in region V , then Stokes’ theorem suggests that this condition will
also guarantee path independence, since for any surface S inside region V a zero curl implies
a zero line integral on the left hand side of (5.83).
One more detail is still needed, however. Stokes’ theorem requires that the integrand
in the surface integral be bounded. This means that in the chosen region V , no singularities
of the integrand can be present. In relation to the closed path integral, this kind of region
is formally referred to as a simply connected region.
c 2006 Tomas Co, all rights reserved
° 179

Definition 5.2 A region V is simply connected if any simple closed path in the region can
be continuously deformed to a single point. Otherwise, the region is multiply connected.

A set of example regions are shown in Figure 5.15. The first case is shown in Fig-
ure 5.15(a) where the region is a solid rectangular box. In this case, we see that any closed
path contained in the region can be deformed continuously into a point inside the box. The
second case is shown in Figure 5.15(b). For this case, a cylindrical subregion has been re-
moved from the center. Even though we could find some closed paths that could deform
to a point, the existence of at least one closed path that will not continuously deform to a
point is sufficient to have the region be classified as multiply connected. The third case is
shown in Figure 5.15(c). In this case, the rectangular box region has a spherical subregion
removed. However, unlike case (b), any closed path inside the region of case (c) can still
be continuously deformed to a point. Thus this third case is also considered to be a simply
connected region.

Figure 5.15: Examples of simply and multiply connected regions. (a) Solid rectangular
region: simply connected (b) Solid rectangular region with a cylindrical subregion removed
from the center: multiply connected (c) Solid rectangular region with a spherical subregion
removed from the center: simply connected.
180 c 2006 Tomas Co, all rights reserved
°

We can now summarize the preceding discussion in the following theorem:

Theorem 5.3 Let F be a vectorR field whose curl is equal to zero inside a simply connected
region V . Then the integral, C F · dR, is independent of path inside the region V .

Example 5.6

Consider the following vector fields:

H = y δ x + x δ y + 2z δ z
G = x2 + y 2 + z 2 δ x + 2xy δ y + (y − z) δ z
y+2 x−1
F = − δx + δ + 2 δz
2
(x − 1) + (y + 2) 2 (x − 1)2 + (y + 2)2 y

and the following paths with the parameter t ranging from 0 to 1:

Path C1 : x = 3 + cos(3πt)
y = 3 + sin(3πt)
z = 2t

Path C2 : x = −12t2 + 10t + 4


y = 3 + 4t(1 − t)
z = 2 − 2(1 − t)2

Path C3 : x = −6t2 + 5t + 1
y = 2t2 − 5t
z = −6t2 + 7t

Path C4 : x = 10t2 − 11t + 1


y = 2t2 − 5t
z = 10t2 − 9t

Figure 5.16 shows paths C1 and C2 . Note that C1 has a helical form while C2 is
a simpler three dimensional curve. Figure 5.17 shows paths C3 and C4 together.
Included in the figure is a line described by: (x, y, z) = (1, −2, z). This line
includes the singular point of vector field F. Thus the line is only relevant when
considering the F.
c 2006 Tomas Co, all rights reserved
° 181

Figure 5.16: Paths C1 and C2 .

Table 5.1: Line integrals based on different vector fields and paths.

Paths
Start: (4,3,0) Start: (1,0,0)
Vector Fields End: (2,3,2) End: (0,-3,1) Curl
C1 C2 C3 C4
H -2 -2 1 1 0
G -36.062 -45.2 -4.4 -11.333 −δ x + 2xδ y
F 4.343 4.343 -1.927 4.356 0

We
R can tabulate
R the results
R of calculating the various type of line integrals of
C
H · dR, C G · dR and C F · dR for different paths. This is shown in Table 5.1
together with curl of the respective vector fields.
As expected, the line integrals for H are path independent because the curl is
zero and the whole three dimensional space is simply connected. For G, since the
curl is not zero, the line integrals depend on the path. Finally, for F, since there
exists a simply connected region that covers C1 and C2 , the two line integrals are
expected to be equal. However, as we increase the region in order to contain paths
C3 and C4 , the allowed region (after removal of regions containing singularities)
are multiply connected. Thus, the theorem states that we expect that path
independence is no longer guaranteed even though the curl along paths C3 and
C4 are still zero.

♦♦♦
182 c 2006 Tomas Co, all rights reserved
°

Figure 5.17: Paths C3 and C4 . (Also included here is the line (1, −2, z).

5.8 Leibnitz Derivative Formula


Theorem 5.4 Given a function F (α, x) that is differentiable in x and α, then

h(α) h(α)
d ∂F (α, x)
Z Z
F (α, x) dx = dx
dα g(α) g(α) ∂α
∂h(α) ∂g(α)
+F (α, h(α)) − F (α, g(α)) (5.84)
∂α ∂α

Proof:

Using the definition of a derivative:


ÃZ ! ÃZ
h(α) h(α+∆α)
d 1
F (α, x) dx = lim F (α + ∆α, x)dx
dα g(α) ∆α → 0 ∆α g(α+∆α)
Z h(α) !
− F (α, x)dx (5.85)
g(α)

The first integral in the left hand side of (5.85) can be divided into three parts,
Z h(α+∆α) Z h(α+∆α)
F (α + ∆α, x) dx = F (α + ∆α, x) dx
g(α+∆α) h(α)
c 2006 Tomas Co, all rights reserved
° 183

Z h(α)
+ F (α + ∆α, x) dx
g
Z (α)
g(α)
+ F (α + ∆α, x) dx (5.86)
g(α+∆α)

We will now approximate the first integral in the left hand side of (5.86) using
the trapezoidal rule:
Z h(α+∆α)
1£ ¡ ¢
F (α + ∆α, x) dx ≈ F α + ∆α, h(α+∆α)
h(α) 2
¡ ¢¤ ¡ ¢
+ F α + ∆α, h(α) h(α+∆α) − h(α) (5.87)
Likewise, we can also approximate the third integral in the left hand side of (5.86)
as
Z g(α)
1£ ¡ ¢
F (α + ∆α, x) dx ≈ F α + ∆α, g(α+∆α)
g(α+∆α) 2
¡ ¢¤ ¡ ¢
+ F α + ∆α, g(α) g(α) − g(α+∆α) (5.88)

Substituting (5.86) (5.87) and (5.88) into (5.86),


Z h(α) "Z
h(α) µ ¶
d F (α + ∆α, x) − F (α, x)
F (α, x) dx = lim dx
dα g(α) ∆α→0 g(α)) ∆α
¡ ¢ ¡ ¢
F α + ∆α, h(α+∆α) + F α + ∆α, h(α) ¡ ¢
+ h(α+∆α) − h(α)
¡ 2∆α¢ ¡ ¢
F α + ∆α, g(α+∆α) + F α + ∆α, g(α) ¡ ¢
+ g(α) − g(α+∆α)
2∆α
Z h(α)
∂ dh dg
= F (α, x) dx + F (α, h(α)) − F (α, g(α)) (5.89)
g(α)) ∂α dα dα

QED

Theorem 5.5 Given a function f (x, y, z, α) that is differentiable in an open


region containing three dimensional region V (α) and where both function
f (x, y, z, α) and region V (α) are both differentiable with respect to the parameter
α. Then,
d ∂f ∂R
Z Z Z
f (x, y, z, α)dV = dV + f (x, y, z, α) n · dS (5.90)
dα V (α) V (α) ∂α S(α) ∂α

where S(α) is the bounding surface of volumetric region V (α), n(x, y, z) is the
unit outward normal vector at the point (x, y, z) on surface S, and R = xδ x +
yδ y + zδ z is the position vector.
184 c 2006 Tomas Co, all rights reserved
°

Proof:

From the definition of the derivative,


·
d 1
Z Z
f (x, y, z, α)dV = lim f (x, y, z, α + ∆α)dV
dα V (α) ∆α→0 ∆α V (α+∆α)
Z ¸
− f (x, y, z, α)dV (5.91)
V (α)
R
By adding and subtracting the term V (α) f (x, y, z, α + ∆α)dV in the right hand
side,
·Z
d 1
Z
f (x, y, z, α)dV = lim f (x, y, z, α + ∆α)dV
dα V (α) ∆α→0 ∆α V (α)
Z ¸
− f (x, y, z, α)dV
V (α)
·Z
1
+ lim f (x, y, z, α + ∆α)dV
∆α→0 ∆α V (α+∆α)
Z ¸
− f (x, y, z, α + ∆α)dV
V (α)
∂f
Z
= dV
V (α) ∂α
·Z
1
+ lim f (x, y, z, α + ∆α)dV
∆α→0 ∆α V (α+∆α)
Z ¸
− f (x, y, z, α + ∆α)dV
V (α)
(5.92)

The last group of terms in the right hand side (5.92) is the difference of two
volume integrals involving the same integrand. We can combine these integrals
by changing the volume of integration to be the region between V (α + ∆α) and
V (α).
Z Z
f (x, y, z, α + ∆α)dV − f (x, y, z, α + ∆α)dV =
V (α+∆α) V (α)
Z
f (x, y, z, α + ∆α)dV (5.93)
V (α+∆α)−V (α)

We could approximate the differential volume in (5.93) as the parallelepiped


formed by the vectors (∂R/∂u)du, (∂R/∂v)dv and (∂R/∂α)dα, where u and v
are parameters used to describe surface S(α). This is shown in Figure 5.18.
c 2006 Tomas Co, all rights reserved
° 185

Figure 5.18: Graphical representation of differential volume emanating from points in S(α)
towards S(α + ∆α).

Recall that µ ¶ µ ¶
∂R ∂R
du × dv = n dS
∂u ∂v
which then gives a differential volume attached to S(α)
µ ¶
∂R ∂R ∂R
dV |(x,y,z)∈V (α+∆α)−V (α) = · × dαdudv
∂α ∂u ∂u
∂R
= · n dα dS (5.94)
∂α
The volume integral for points bounded between the surfaces of V (α) and V (α +
∆α) can now be approximated as follows:
Z
f (x, y, z, α + ∆α)dV
V(α+∆α) −V(α)
∂R
Z
≈ f (x, y, z, α + ∆α) · n ∆α dS (5.95)
S(α) ∂α

Substituting (5.95) into (5.93) and then to (5.92),


d ∂f
Z Z
f (x, y, z, α)dV = dV
dα V (α) V (α) ∂α
1 ∂R
Z
+ lim f (x, y, z, α + ∆α) · n∆αdS
∆α→0 ∆α S(α) ∂α
∂f
Z
= dV
V (α) ∂α
186 c 2006 Tomas Co, all rights reserved
°

∂R
Z
+ f (x, y, z, α) · n dS (5.96)
S(α) ∂α

which is the Leibnitz rule for differentiation of volume integrals.

QED

You might also like